veritas-kanban/CHANGELOG.md

Changelog

All notable changes to Veritas Kanban are documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

Unreleased

6.1.1 - 2026-08-22

Veritas Kanban 6.1.1 restores reliable Task Detail scrolling after the Mantine tabs migration and completes a security-audited dependency maintenance pass.

Changed

• Clarified that independent review is owner-directed and optional rather than a default delivery SOP or release gate (#1156).
• Updated supported runtime and development dependencies across the workspace, including Chalk 6, and refreshed transitive security override floors so both production and full dependency audits resolve without known vulnerabilities (#1149, #1155).
• Deferred jsdom major updates in Dependabot while the current major requires a higher Node.js patch floor and breaks the Mantine-backed web test environment; jsdom 29 patch updates remain enabled (#1148).

Fixed

• Restored the shared overlay flex-column contract so long Task Detail content remains height-constrained and scrollable. Added a Chromium regression that verifies layout, overflow, and real wheel scrolling through the drawer (#1153, #1154).
• Prevented interactive controls inside task cards from activating the card, restoring reliable touch status selection in WebKit after the Mantine 9.5 update. Status-move browser coverage now waits for the visible save contract before closing Task Detail (#1156).
• Made file-backed workflow operations wait for their storage directory to be ready, eliminating a startup race that could return ENOENT when the first workflow request arrived immediately after service construction (#1156).

6.1.0 - 2026-07-26

Added

• Added complete workspace knowledge collections: immutable classified sources, cited and versioned derived pages, reviewed ingestion dry runs, atomic apply and reversal, cited keyword/QMD search, query promotion, and cited work-product export. Run launch evidence now restricts every agent read and mutation to the exact shared source and page resources in the persisted manifest, isolates QMD projections by manifest digest, and binds promoted or exported search evidence to the same run. Confidential and restricted previews withhold content and force redacted exports even when the collection otherwise allows full output. File and SQLite storage preserve the same workspace, digest, attribution, idempotency, contradiction, graph, and activity contracts (#867).
• Added launch-scoped deterministic integrity linting for knowledge collections. Stable, digest-bound findings detect structural graph errors, invalid schemas and metadata, provenance gaps, changed source hashes, invalid citation locations, configurable freshness violations, orphan pages, missing canonical terms, and unanswered research questions without retaining source text in findings (#868).
• Added explicit material-claim lifecycle controls for knowledge collections. Compare-and-set transitions are versioned, actor-attributed, evidence-linked, digest-bound, idempotent, and reversible through page history. Agents can flag claims as needs-review or disputed while only administrators can finalize supersession, retraction, or resolution, so conflicting claims remain visible instead of being silently overwritten (#868).
• Added durable knowledge-integrity operations across file and SQLite storage. Bounded cursor-resumable lint chunks persist idempotently by run identity; optional semantic candidates flag contradictions, near-duplicates, supersession, and evidence gaps with both sides' page, claim, and source identities. Findings retain owner, status, severity, acknowledgement, due date, remediation links, occurrence history, and compare-and-set transitions, while a launch-scoped health endpoint exposes last observation, overdue work, and actionable status for scheduled workflows and operators (#868).
• Added preview-first, attributable workspace checkpoint rewind for run-owned worktrees. Immutable content-addressed checkpoints capture Git, index, files, exclusions, ownership, conversation cursors, exact provider hunk ranges where explicit unified diffs exist, bounded retention, and direct comparisons; conflict-aware previews support digest-bound per-path accept, reject, and leave-untouched decisions, while recoverable storage transactions mutate only selected paths and preserve descendant state on failure. The production control route now quiesces an exact active Codex app-server turn and forks an earlier approved turn into a new live provider thread, while ambiguous cursors, external edits, unsupported providers, stale runtime evidence, and unresolved non-attribution conflicts fail closed (#872).
• Added durable execution-tree cancellation and a provider-neutral fan-out circuit breaker. Operators can cancel one queued launch or an entire root objective through REST, vk admission, and Operations; root cancellation is recorded before descendants drain so late resume, retry, fallback, workflow-step, and child-agent launches fail closed. The breaker evaluates durable descendant count, depth, active and queued work, capacity pressure, and aggregate budget pressure before every expansion, persists bounded evidence across restart, and requires a safe operator resume before the tree can grow again (#1055).
• Routed scheduled workflows, watcher continuations, conversation resume and follow-up, retries, fallbacks, provider handoffs, recovery, and child-agent starts through the same durable admission and queue contract used by direct tasks and workflow steps. Every source preserves its causal execution-tree identity, revalidates immutable launch evidence before dispatch, and rejects hidden adapter-owned queues or bypass launches (#1054).
• Added one bounded durable admission queue shared by direct tasks and workflows. Atomic dequeue claims both a queue lease and capacity; restart recovery expires abandoned claims without duplicate dispatch. The scheduler combines explicit priority, capped age promotion, workspace fairness, and deterministic tie-breaking while retaining redacted conditional selection evidence. REST, vk admission, and Operations expose queue health, readiness, limiting scopes, age, priority, and safe cancellation with file and SQLite parity (#1053).
• Added versioned execution-tree identities and aggregate budget reservations to direct attempts, conversation continuations, retries, fallbacks, provider handoffs, child agents, workflow roots, and workflow steps. Capacity and the strictest applicable workspace, agent, workflow, run, and root-objective budgets are now claimed atomically through the same file lock or SQLite transaction. Idempotent usage events convert reservations into attributable committed input/output/total tokens, cost, tool calls, runtime, idle time, retries, and fan-out without counting descendant totals again. Terminal release preserves committed usage and returns unused reservations. REST and vk admission tree expose bounded totals, remaining policy capacity, contributors, and the exact blocking policies (#1052).
• Routed workflow roots and every provider-backed workflow step through the durable admission controller. A root now reserves workflow-control capacity before the run becomes active, while each executable step reserves against its resolved provider and selected host before step-attempt mutation or adapter dispatch. Retry and fallback attempts use new child reservations only after the prior reservation releases. Terminal, cancellation, and restart-reconciliation paths release idempotently, and restart recovery reclaims only the exact persisted root binding. Workflow run, step, and root-reservation filters are available through REST and vk admission; file and SQLite storage preserve the same inspection contract (#1051).
• Added durable admission reservations for every direct agent task launch. File and SQLite backends atomically enforce one active run per task plus optional global, workspace, root-task, provider, and host ceilings for run slots, process slots, and operator-estimated memory. Versioned decisions distinguish retryable overload from terminal policy denial before attempt persistence or provider dispatch. Lease recovery is bound to verified live supervisors, terminal paths release idempotently, and file heartbeat history compacts atomically. Operators can inspect active and recent reservations through read-scoped REST and vk admission JSON commands (#1050).
• Added a first-class workflow definition browser with deep-linked view, edit, and duplicate routes. Definition detail now exposes agents, ordered steps, phases, inputs, acceptance criteria, gates, loops, parallel branches, outputs, and provenance before execution. Server-owned access evidence distinguishes user-owned, shared, and built-in workflows; read-only definitions explain the restriction and offer duplication when permitted. User-owned edits save through Author with version-bound conflict protection that preserves the draft on failure. Starting a run is now a separate configuration step for optional task association and JSON context (#940).
• Enforced active phase authority across run tool discovery, mediated invocation, approvals, completion evidence, REST, CLI, and the task run timeline. MCP readOnlyHint annotations now classify external reads while unannotated tools fail closed as mutations; disallowed tools and credentials stay out of launch catalogs, and hidden or stale calls are rejected against the current transition evidence before dispatch. Approval decisions cannot outlive or widen their bound phase, and completion records identify the effective phase plus every authority-expanding transition. Redacted support bundles include bounded phase identities, authority counts, source kinds, transition expansions, and completion bindings without exporting exact paths, credentials, or full digests. ACP stdio exposes the required pre-execution mediation; adapters without equivalent command and external-action controls return typed blockers for explicit phases. Legacy attempts remain readable without invented transition state (#1033).
• Propagated immutable phase authority through task previews and starts, workflow steps, retries and fallbacks, provider changes, conversation resume, follow-up and fork operations, and active-run controls. Every executable entry now resolves the exact parent launch or current transition evidence before attempt mutation, persists the compiled phase digest and source references, intersects descendants monotonically, and returns typed blockers when the selected sandbox, runtime, host, or tool policy cannot enforce a required phase dimension. Legacy launches remain explicit and readable without inventing restrictions (#1036).
• Added the append-only phase-transition-record/v1 journal for active runs. Every transition is bound to the exact attempt, prior sequence, compiled phase evidence, and launch-manifest digest. Narrowing transitions apply immediately, while authority expansion uses the existing exact-action approval broker. Administrators can apply an auditable emergency expansion for at most 24 hours; expiry durably restores the prior evidence. File and SQLite repositories preserve restart recovery and idempotency, and new REST and CLI controls expose current state, history, transitions, and approval decisions (#1035).
• Added the provider-neutral phase-capability-profile/v1 foundation with strict schemas and built-in explore, plan, implement, verify, and publish profiles. A pure compiler now intersects phase requests with parent, agent, sandbox, tool-catalog, and launch-policy authority without ever widening a scope. Required unsupported or unenforceable dimensions return typed fail-closed blockers, legacy launches are explicit, and the optional plan artifact is restricted to one normalized harness-owned exact path that cannot grant general filesystem or indirect shell-write authority (#1034).
• Added a provider-neutral workspace execution trust gate that scans the exact task worktree for repository-controlled agent instructions, provider configuration, MCP servers, hooks, language-server settings, workflows, extensions, skills, and agent definitions before launch. Stable identity combines canonical worktree, repository, Git common-directory, and credential-redacted remote evidence so sibling, nested, moved, and linked worktrees cannot borrow authorization. Append-only trusted, restricted, denied, revoked, and expiring decisions bind to an exact inventory digest; project policy can only narrow them. Executable configuration requires explicit authorization, while model-only instructions can run provisionally only under enforced read-only, no-network, credential-free restricted controls. The immutable launch manifest records redacted inventory and decision evidence, and a final pre-spawn rescan blocks any drift. Added administrator REST and CLI scan, decide, and revoke controls (#878).
• Added run-scoped filesystem sandbox enforcement for local ACP, Claude Code, Codex app-server, Codex CLI, and Hermes processes. Required presets compile explicit read, write, deny, dotfile, protected-metadata, temporary, and cache rules into a version-bound codex sandbox wrapper before provider creation. Workspace and home aliases fail closed on traversal or symlink escape, and nested mount boundaries and pre-existing external hard-link aliases are denied and rechecked before spawn. Failed or byte-changed backends block launch. CLI package/virtual-environment roots, linked-worktree Git metadata, and .git, .agents, .codex, and .veritas-kanban directly beneath writable roots are bound read-only, while protected paths cannot themselves be writable roots or symlinks. Ambient Git config is replaced with a run-scoped identity-only environment. Immutable manifests retain only canonical path hashes, the exact provider-runtime digest, backend executable-content and conformance evidence, and durable cleanup state. Local native and wrapped runs receive supervisor-owned temporary and cache directories, while remote native enforcement must prove every active filesystem and lifecycle capability. Cleanup rejects symlinked ancestors (#862).
• Added one durable run-recovery/v1 state machine for production task attempts and workflow steps. Only classified transient transport, provider availability, rate-limit, timeout, and verification failures retry; policy, configuration, cancellation, destructive partial-side-effect, and unknown failures fail closed or require operator review. Recovery records preserve causal parents, exponential jittered backoff, selected routes, launch manifest digests, and cumulative budgets; compatible fallbacks are probed through runtime capability and sandbox gates before launch. Pending recovery survives restart, duplicate terminal callbacks cannot create multiple branches, and exact task recovery can be cancelled through REST, CLI, or MCP controls. Workflow recovery exposes the same exact-parent cancellation through REST (#861).

Changed

• Aligned contributor verification with the sustainable delivery cadence: each change now gets one risk-proportional review, focused checks for the affected product boundary, and runtime smoke tests only when runtime behavior changes. Complete workspace build, typecheck, test, security, integration, E2E, and artifact gates remain milestone-only, while the dependency-free cadence checker now rejects the retired per-commit multi-review, per-merge broad-gate, and unconditional browser-smoke language if it returns. Cadence checker changes are verified inside the CI scope-control job and no longer trigger an unrelated workspace unit suite (#1048).
• Standardized focused Vitest verification on direct pnpm --filter <package> exec vitest run <exact-test-files> invocation. The dependency-free delivery cadence checker now rejects active guidance that presents either of the ambiguous package test -- <test-files> or test -- --run <test-files> wrappers as focused verification, preventing an intended file slice from silently expanding into an entire package suite (#1044, #1058).
• Reworked GitHub release notes to use natural page-width prose and concise lists instead of ragged hanging-indent blocks or unmarked stacks of bold-led paragraphs. The release-format gate now rejects long wrapping list items, nested headings, bold-led prose items, and consecutive sentence-sized blocks in v6.0.2 and later release bodies (#1025).

Fixed

• Completed append-only admission snapshot writes before sync so a short FileHandle.write() cannot leave a truncated JSONL record that makes durable reservation state unreadable (#1139).
• Restored the complete node:fs/promises surface in filesystem test doubles after centralized storage added lstat, clearing collateral concurrent-suite failures without weakening production path checks (#1136).
• Mapped the exact knowledge-collection router prefix to shared work-product read/write permissions, restoring fail-closed client/server permission coverage parity (#1141).
• Updated the CLI native-fork regression coverage to validate the required generated idempotency key before comparing the stable request payload, clearing the release-branch related-test gate (#1143).
• Separated live MCP-to-HTTP task and sprint integration groups from the server-independent unit suite so focused CI no longer attempts localhost API calls without a running server (#1144).

6.0.2 - 2026-07-24

Veritas Kanban 6.0.2 is a desktop recovery and supportability hotfix. It keeps Chat contained within a reversible Workbench dock, adds authoritative native version/build information, and makes CI verification proportional to the change while retaining full release gates.

Changed

• Made CI test scope deterministic and path-aware. Documentation-only changes skip workspace suites, ordinary code changes run Vitest related coverage for affected packages, and CI, manifest, shared, storage, or desktop changes conservatively select the full gate. A successful reviewed full-suite head is reused after merge only when it is an ancestor of the merge commit, avoiding a duplicate full run while scheduled and explicit release gates remain authoritative. Dual-storage parity now invokes its exact Vitest file instead of expanding through the package test wrapper, and job summaries retain exact range and selection evidence (#1000).

Fixed

• Updated QMD query scoping to pass one documented -c flag per collection. Current QMD versions silently ignore the former unknown --collections argument, which could broaden a supposedly scoped search (#867).
• Replaced the desktop-only bottom Chat surface with a bounded dock that opens on the right by default and can switch between Right and Bottom without remounting the active conversation. Both dimensions are clamped against the live viewport, chat scrolling cannot move the application shell, wheel input cannot resize the dock, and Close, Escape, Back, or Reset Layout always preserve a visible board and restore focus (#1004).
• Added a native About Veritas Kanban panel and offline Copy Version Information action using Electron's authoritative application version. The same record now reports the embedded release commit, stable/beta/development channel, OS, and architecture through About, clipboard support text, the desktop bridge, and updater fallback without depending on the renderer or network (#1005).

6.0.1 - 2026-07-24

Veritas Kanban 6.0.1 is the first supported stable v6 release. It supersedes the quarantined 6.0.0 prerelease after closing the post-publication desktop and workflow stabilization backlog tracked in #924.

Fixed

• Kept desktop Chat in a bounded, reversible workbench panel; Escape and browser Back now close it, obsolete persisted panel visibility is discarded, compact window heights preserve the application shell, and the native Navigate menu can reset layout state without deleting application data (#945).
• Normalized omitted workflow and run collections in the task Workflow action, preventing undefined length crashes and providing a recoverable load error with Retry instead of replacing the task surface with an error boundary (#936).
• Made New Profile open a focused, validated scoring draft from both Profiles and Score Explorer, with cancel returning to the originating tab without creating an orphan profile (#943).
• Let Archive cards grow with expanded content, clamp collapsed text by whole lines with accessible full text, wrap long metadata safely, and keep Restore reachable beside long task content (#939).
• Established one bounded, keyboard-focusable scroll owner for task drawers and shared overlays, kept drawer chrome fixed, and made task description editors taller and vertically resizable (#935).
• Preserved in-app route origins, scroll positions, and task return paths across full-page navigation, added browser Back and Cmd+[ support, and made direct links fall back safely to Board (#937).
• Reconciled Operations Digest totals with board inventory, exposed auditable inclusion and exclusion reasons plus metadata quality findings, bounded observed wall time to the selected window, and documented current-state versus windowed metrics (#944).
• Removed competing profile-editor and scorer-list scroll containers so Agent Output Scoring uses the application page scrollbar at compact and full desktop sizes (#938).
• Expanded the template editor into a responsive authoring surface with one bounded scroll region, fixed actions, a resizable Markdown editor, inline validation, and unsaved-change protection (#941).
• Made first-run setup version-neutral and sourced desktop bridge version reporting from Electron's packaged application metadata (#986).

6.0.0 - 2026-07-24

Quarantined prerelease. Use 6.0.1 or newer. The 6.0.0 artifacts remain available only for investigation and historical release evidence.

Each release entry below names its tracking issue. The complete issue-to-pull request mapping, including multi-PR fixes and release engineering, is retained in the v6 Release Candidate Evidence Packet.

Changed

• Tiered GitHub Actions so pull requests run changed test files with lint/typecheck, security, and build gates, while the complete workspace suite runs on main, nightly, by manual dispatch, or with the ci:full label. Cross-platform unsigned desktop packaging now runs for desktop/package PRs, after relevant changes merge to main, or by manual dispatch (#955).

Fixed

• Prevented the packaged desktop updater from treating an older stable release as downloadable. Stable, beta, and dev channel assignment now explicitly restores allowDowngrade=false after the electron-updater channel setter changes it internally (#983).
• Deferred the default run-supervisor SQLite repository lookup until the first supervisor operation. Packaged SQLite startup can now import and initialize routes before storage bootstrap without crashing (#981).
• Added bounded macOS desktop startup verification that retries through the LaunchServices readiness window, requires the installed app and health endpoint versions to agree, and reports process, port, and log diagnostics on a real timeout. Updated web-to-desktop migration, routine Homebrew upgrade, automation, troubleshooting, and release guidance to use the readiness gate instead of an immediate health request (#927).

Added

• Added a composed credential-free Buzz compatibility gate and complete machine-readable seam evidence for the pinned Buzz 0.4.24 / 710ed9fff57878a1d69f809b80a6ee0416c53fc4 baseline, plus trigger setup, support dispositions, optional live-smoke boundaries, upgrade, and rollback guidance (#912).
• Added allowlisted Buzz root-message workflow triggers with bounded predicates, provider-neutral pre-dispatch hooks, durable causal keys, replay and echo suppression, restart reconciliation, API configuration, and audit history (#911).
• Added the provider-neutral runtime-hook/v1 in-process hook bus with bounded envelopes, deterministic scope ordering, explicit blocking pre-events, passive post-events, timeout/reentrancy protection, side-effect-free dry-run, and causal run-journal evidence. External executable and HTTP handlers remain fail closed pending their sandbox and egress boundaries (#874).
• Added harness-compatibility-matrix/v1 for Buzz, Grok Build, OpenAI Codex app-server, Claude Code, and GitHub Copilot CLI. One reviewed record now drives exact builds, source caveats, capability and fixture digests, invalidation policy, live support tiers, API, vk doctor, Settings, telemetry, and operator guidance (#918).
• Added harness-conformance-suite/v1 and a deterministic runner for seeded, repeated provider/model/profile/policy/sandbox comparisons. Explicit assertions cover outcomes, files, tools, approvals, network, policy, completion, and capability evidence; normalized results aggregate pass rate, variance, latency, tokens, cost, retries, failure classes, immutable evidence references, and versioned baseline regressions. A credential-free recorded mock command provides the CI/local foundation while credential-gated lanes require explicit opt-in (#859).
• Composed the pinned buzz-agent ACP profile with the system-owned veritas-run bridge. Buzz sessions now receive only that bridge for selected run tools, so catalog allow, deny, approval, attribution, and cleanup remain under the provider-neutral control plane without native server credentials or a Buzz-specific MCP implementation. Provider capability evidence advances to probe revision 14 (#909).
• Added the system-owned veritas-run MCP bridge for credential-bound tools. Opaque, in-memory authority binds the exact task, attempt, catalog, launch manifest, and two allowed bridge methods. Codex CLI/SDK, Codex app-server, Claude Code, and ACP stdio inject the same narrow bridge; Hermes and OpenClaw fail closed before dispatch when a credential-bound catalog is selected (#970).
• Added exact-action credential lease consumption to mediated run-tool calls. The server binds each operation to the active launch manifest, catalog, server, tool, arguments digest, approval, and operation ID; resolves values only inside a one-shot downstream MCP session; rejects replay, drift, approval mismatch, unavailable sources, and credential-bearing results; and relies on the existing terminal lifecycle reconciliation to revoke leases (#969).
• Added value-free credential boundary evidence to run-tool-catalog/v1. Credential-bound definitions now require enabled MCP-scoped broker definitions, exact source targets, and immutable definition/scope digests. Discovery strips source environment and header values, native provider configuration omits credential-bound servers, covered launch references report a tool-control-plane boundary, and uncovered or drifted references remain fail closed (#968).
• Added vk acp serve --stdio and vk acp status --json to expose a task-bound, provider-neutral ACP v1 server view over Veritas-managed conversations. Fresh prompts retain the immutable task envelope, reconnect replays the causal run journal by cursor, permission requests use the durable approval broker, cancellation interrupts instead of stopping the task, and disconnect leaves the supervised run active. Client-owned MCP catalogs and mismatched worktrees fail closed (#960).
• Added a disabled-by-default Grok Build profile under the generic ACP provider, pinned to released v0.2.111 build 94172f2aa4e5. Veritas now compiles a dedicated no-leader stdio launch, accepts only bounded model, effort, deny, tool, sandbox, and feature-disable controls, classifies Grok boot credentials separately from task credentials, and validates the exact version, capability, and xAI extension handshake. Unknown builds, approval bypass, reauthentication, leader/plugin/endpoint injection, and source provenance claims fail closed. Provider probe evidence advances to revision 13 (#920).
• Added a disabled-by-default GitHub Copilot CLI profile under the generic ACP provider, pinned to the Copilot 1.0.74 public-preview handshake. Veritas now owns a safe stdio baseline, compiles bounded restrictive model/effort/ tool/URL/context/credit settings into immutable launch argv, classifies Copilot boot credentials without probing authentication or inference, and reports the incomplete-source and release-tag provenance limits truthfully. Provider probe evidence advances to revision 12 (#917).
• Added a disabled-by-default buzz-agent conformance profile under the generic ACP provider, pinned to Buzz v0.4.24 commit 710ed9fff57878a1d69f809b80a6ee0416c53fc4. The profile validates exact ACP identity, version, capabilities, known limitations, safe configuration and boot-authentication environment keys, rejects buzz-acp and incompatible capability drift, and advances provider probe evidence to revision 11 (#907).
• Added run-launch-credential-plan/v1 to classify provider boot authentication, broker-targeted task credentials, and explicit high-risk environment passthrough in immutable launch evidence. Plans bind to provider runtime evidence, remain value-free, ignore timestamp-only probe refreshes, and block task credential launches until an accepted tool or egress boundary proves brokered delivery (#932).
• Added a generic stable ACP v1 stdio provider adapter with no-shell supervised launch, initialize-time identity and capability evidence, fresh/resume/fork/ close session lifecycle, bounded bidirectional JSON-RPC, causal message, thought, plan, and tool events, durable permission brokering, capability drift rejection, explicit environment filtering, and immutable all-allow run-scoped MCP session configuration. Provider capability evidence advances to probe revision 10; the client-facing ACP server view is tracked separately by #960 (#870).
• Added the tool-server-definition/v1, tool-server-discovery/v1, and run-tool-catalog/v1 control-plane contracts with file and SQLite persistence, version-bound discovery caching, required and optional server posture, per-run allow/deny/approval decisions, bounded supervised stdio and Streamable HTTP invocation, causal replay-safe events, REST/CLI/MCP operator surfaces, and native run-scoped MCP injection for Codex app-server and Claude Code. Credential-bound definitions remain fail-closed until the provider launch credential broker lands (#857).
• Added conversation-lifecycle/v1 with durable provider conversation, turn, item, parent, and fork identities; launch-manifest and worktree compatibility checks; context-window posture; causal lifecycle events; attributed REST, CLI, MCP, and web controls; and explicit unsupported delivery results. Codex CLI and SDK can resume exact persisted sessions, Claude Code can resume or fork with system-owned native flags, and Codex app-server can resume, follow up, steer, fork, compact, archive, interrupt, and close through schema-validated JSON-RPC methods. Generic process stdin is no longer treated as a successful provider follow-up path, and capability evidence advances to probe revision 8 (#856).
• Added run-supervisor/v1, a durable provider-run ownership record with exact provider, task-envelope, launch-manifest, and worktree bindings; persisted process groups or remote-session handles; budget and event cursors; expiring single-owner leases; typed recovery actions; and idempotent terminal ownership. Startup now reattaches verified local provider processes, resumes journal reconciliation strictly after the durable cursor, restores exact completion results across crash races, and blocks unsafe recovery without signaling reused PIDs or starting duplicate work. File and SQLite repositories share the same compare-and-set contract, and provider capability evidence advances to probe revision 7 (#853).
• Added a provider-native run-approval/v1 broker with file and SQLite persistence, exact action hashes, atomic compare-and-set decisions, auth-derived reviewer identity, critical-action authentication freshness, expiry/cancellation, mobile two-layer safety, causal audit correlation, REST and WebSocket delivery, task and shared-run review surfaces, and non-persisted elicitation response values. Codex app-server now pauses and resumes its exact command, file, permission, tool-question, legacy exec/patch, and MCP elicitation request through pinned v0.145.0 response schemas; stale, changed, duplicate, interrupted, unsupported, and failed requests deny safely (#852).
• Added a first-class Codex app-server v2 task adapter pinned to codex-cli 0.145.0 and upstream commit 25af12f7e61572b0bc18ddb1008be543b91519b0. Veritas now owns a strict-stdio, no-shell launch, validates checked-in exact-version JSON Schemas, persists one task-bound thread and turn, streams item/usage/completion events into the causal journal, supports cooperative turn/interrupt, and retries only bounded overload -32001 responses. Inherited MCP, hooks, plugins, apps, browser/computer tools, elicitation, remote control, and unsandboxed thread/shellCommand remain structurally disabled; approval, lifecycle, and reattachment controls fail closed until their provider-neutral brokers land (#921).
• Added a first-class Claude Code v2.1.218 task adapter with a reproducible no-shell bare-mode launch, static sandbox-derived permissions, explicit environment authentication with provider credentials scrubbed from tool subprocesses, bounded health and agent-discovery diagnostics, immutable task/run manifests, drain-safe stream-json ingestion, causal event mapping for partial text, thinking, tools, hooks, subagents, usage, cost, and results, separate session identity persistence, artifact discovery, and fail-closed capability evidence for lifecycle and broker controls that have not landed yet. Legacy provider-less Claude records migrate only when command identity matches, and the old permission-bypass default is removed (#916).
• Added run-event/v1, a provider-neutral causal event journal for OpenClaw, Codex CLI, Codex SDK, Codex app-server, Claude Code, and Hermes. Provider mappers now durably append bounded, redacted, deduplicated lifecycle, message, reasoning, command, file, tool, approval, artifact, usage, error, and unknown events before projecting them into legacy logs, traces, telemetry, budgets, and live output. File and SQLite repositories allocate per-attempt monotonic cursors, REST and WebSocket clients can replay after a cursor, and reconnect subscriptions merge replay with concurrent live events without gaps. The published JSON Schema, symlink-safe file storage, SQLite migration, provider fixtures, replay, migration, redaction, and interleaved operator/provider tests define the v1 compatibility boundary (#850).
• Added one-way Buzz persona/team definition import with signed bounded NIP-33 head queries, exact v0.4.24 field policy, preview/diff/collision actions, same-author team resolution, disabled profile and roster materialization, immutable provenance, source-status reporting, optimistic JSON/SQLite mutations, audit history, Settings controls, and malicious-source coverage. Runtime/model/provider preferences remain source-only, and import never launches, enables, routes, fetches, installs, or writes back (#910).
• Added a reference-only Buzz communication adapter with typed configuration, host-preserving HTTP/WS normalization, SSRF-aware bounded relay probes, NIP-11 software/version/community checks, NIP-98 identity and membership/read diagnostics with built-in hexadecimal/nsec signing, stable machine-readable reason codes, evidence invalidation, shell-free optional command discovery, Settings controls, vk doctor reporting, and operator/API/security documentation. The resulting compatibility evidence is the fail-closed gate consumed by Buzz communication delivery (#906).
• Added a native bidirectional Buzz channel bridge for Squad Chat with signed Nostr root/reply events, channel mapping APIs and Settings controls, NIP-42 WebSocket subscriptions, source author/timestamp/deep-link metadata, (created_at,event_id) cursors, overlap dedupe, out-of-order reply recovery, loop prevention, non-destructive edit/delete auditing, delivery-unknown reconciliation, runtime health facets, safe disable/rollback, fake-relay contract coverage, and operator/API/security documentation (#906).
• Added transactional, remote-safe worktree lifecycle management with versioned manifests, exact fetched base commits, reasoned offline fallback, task/attempt ownership leases, active-run locks, recovery states, and preview-first stale cleanup. Dirty, untracked, unpushed, unmerged, and externally held worktrees now require an admin-audited override. Attempt leases use exclusive claims, legacy worktrees have a validated adoption path, and restart recovery reconciles partial rebase, integration, push, and cleanup states. Integration uses a dedicated temporary worktree and non-force push without mutating the primary checkout. Task envelopes and run launch manifests bind the exact worktree allocation and base evidence (#858).
• Added the versioned credential-definition and run-bound lease core with metadata-only admin APIs, canonical definition/scope/action fingerprints, opaque hashed handles, exact launch-manifest and action binding, atomic TTL and use-count enforcement, operation-level replay rejection, refresh/revoke/expiry/reconciliation state, callback-only secret resolution, terminal-run invalidation with revocation retry, live-owner file locking, and bounded metadata audit records. Required brokered presets now reject advisory or externally delegated credential handling until a controlled provider boundary proves end-to-end support (#931).
• Added the immutable run-launch-manifest/v1 contract, a preflight preview API and CLI command, per-field configuration provenance, fail-closed profile tool/MCP/permission/health enforcement, parent-attempt material drift, readiness/override evidence, governance traces, attempt/history/log persistence, and completion-packet links to the exact launch and provider capability evidence (#854).
• Added immutable provider-owned task-envelope transports for OpenClaw, Codex CLI, Codex SDK, Codex app-server, Claude Code, and Hermes. Each renderer binds the exact envelope and runtime identity, commit policy, bounded attributed profile/checkpoint context, workspace baseline, verification gates, and completion evidence to the launched request. OpenClaw alone receives a completion callback; process and stream providers return terminal output through harness-owned capture, and mismatched provider/adapter identities fail closed (#892).
• Added digest-bound, idempotent completion-result/v1 persistence for process, stream, callback, remote-session, and operator-interruption terminal paths. Provider claims are bounded, redacted, and untrusted while harness evidence attributes post-launch files, commits, artifacts, verification, and side effects without crediting pre-existing off-HEAD history. Commit/output/evidence violations downgrade success to recoverable partial completion; exact duplicate callbacks survive retry or restart and conflicting terminal owners fail closed. Provider-owned transports cannot be replaced by callbacks, completion writes use compare-and-swap retries, and harness-owned restart reconciliation remains inside the same result contract while OpenClaw stays eligible for its callback (#893).

Security

• Removed the legacy Claude Code permission-bypass default. Every executable provider now launches through an explicit adapter, immutable launch manifest, bounded environment allowlist, capability evidence, exact-action approval broker, and credential plan. Unsupported controls, adapter/profile drift, stale evidence, and unknown provider-less records fail closed (#919, #916, #852, #932).
• Added value-free credential definitions and one-shot run-bound leases. Credential values are resolved only inside the mediated downstream call and are excluded from provider-native configuration, manifests, logs, telemetry, fixtures, and API responses (#931, #968, #969, #970).
• Added bounded, redacted causal protocol storage and supervised no-shell process lifecycles for ACP, Codex app-server, Claude Code, and Buzz relay integration. The deferred fine-grained HTTP egress proxy remains tracked in #855; any required network rule that current runtime evidence cannot enforce blocks launch.

Migration

• Existing provider configurations are normalized to harness-support-profile/v1. Only known legacy records whose command and built-in type both identify Codex, Hermes, or Claude are migrated to their matching adapter. Provider-less, ambiguous, disabled, or mismatched records remain disabled or fail closed; no migration starts a provider.
• The public API remains v1. New run, approval, lifecycle, tool, credential, compatibility, Buzz, and conformance contracts are additive. Existing v5 board and desktop SQLite data remains in place; operators must back up before upgrade and keep the v5.2.5 backup for rollback.
• Buzz, Grok Build, GitHub Copilot CLI, Claude Code, and Codex app-server support is exact-version and evidence-bound. Unknown builds do not inherit certification. Grok Build and Copilot CLI retain source-provenance limits, ACP lifecycle support varies by runtime, and credential-gated live smoke remains separate from deterministic certification.

5.2.5 - 2026-07-23

Added

• Added a complete web/source-to-Mac migration runbook with a decision path for already-populated desktop SQLite, coherent stopped-writer backups, isolated file-to-SQLite staging, governed bundle replacement, watchdog discovery, authenticated automation cutover, verification, rollback, and warning remediation (#899).
• Added the versioned task-envelope/v1 and completion-result/v1 contracts, canonical envelope digests, immutable per-attempt persistence, launch HEAD and dirty-file attribution with staged-blob and worktree SHA-256 fingerprints, bounded evidence/output/verification/side-effect schemas, and explicit forbidden, allowed, or required commit policy with run, task, and legacy-setting precedence. Baseline capture retries unstable repositories, path side effects are clamped to the effective sandbox, authoritative run contracts reject generic task PATCH mutation, oversized legacy workspace IDs normalize deterministically, and task create/update policy is available over REST, CLI, and MCP. Agent start and status APIs, task storage, attempt history, and logs now carry the exact envelope used for the run (#891).
• Added versioned, evidence-backed provider runtime manifests for Codex CLI, Codex SDK, Hermes, and OpenClaw with bounded identity and conformance probes, race-safe version-skew cache invalidation, canonical immutable run snapshots and digests, complete capability posture, and attempt/history/trace/log persistence (#885).
• Added validated runtime-manifest registration and heartbeat refresh, custom provider discovery, manifest-authoritative host posture, and provider-neutral required-capability routing with structured evidence for every attempted primary, fallback, or rejected candidate (#886).
• Added one provider-runtime control evaluator for launch requirements and status, logs, completion, stop, steer, resume, reattach, tool, MCP, structured output, token-usage, and artifact controls. Agent status now returns concrete capability-derived availability, reasons, and remediation, and REST, CLI, and MCP starts can declare additional required capabilities. Completion callbacks, token budgets, and shared co-drive controls are attempt-bound; provider-native approvals fail closed; OpenClaw task and workflow evidence are separated; and workflow requirements account for standard MCP names and metric-specific budgets (#887).
• Added an admin-governed SQLite journal maintenance workflow with non-mutating previews, restart-time exclusive conversion, verified backups, durable stage journals, forward-only crash recovery, integrity verification, in-place mode rollback while exclusive, single-host ownership locks, expiring/revocable overrides, CLI commands, and redacted diagnostics (#882).

Changed

• Added Hermes to normalized provider profiles and Settings selectors, and made explicitly configured providers without an executable task adapter fail closed instead of silently dispatching through OpenClaw (#885).
• Made host provider, model, tool, and sandbox capability decisions consume validated manifests. Legacy free-form registration posture remains visible but cannot satisfy required runtime capabilities. Runtime evidence writes are identity-bound, reject unredacted secrets and misspelled fields, expire with heartbeat liveness, and invalidate when provider identity changes. Unresolved sandbox presets fail closed instead of borrowing aggregate host signals (#886).
• Made active agent and workflow controls consume the exact persisted launch manifest, reject invalid or mismatched snapshots, and fail closed before unsupported work. Sandbox dry-runs now resolve live registered manifest digests server-side, workflow manifests persist before provider execution, and task, Work view, and shared co-drive controls expose accessible disabled reasons when a runtime cannot stop or steer (#887).
• Classified the authoritative SQLite filesystem before database open, limited WAL to recognized durable local filesystems, refused known-unsafe and unknown storage before creating database sidecars, and exposed redacted filesystem, journal, and integrity posture through deep health and Maintenance diagnostics (#881).

Fixed

• Raised vulnerable transitive fast-uri and js-yaml ranges to patched releases so the production dependency audit passes at the release gate (#903).
• Made startup, setup, password, recovery, and login surfaces draggable in the frameless desktop window, and made first-run setup detect populated desktop SQLite data, default to a non-destructive Use Existing Data path, show concise migrated-data counts, and reserve backup restore for recovery (#901).
• Serialized agent launch and terminalization ownership per task and attempt so concurrent starts, stop requests, provider exits, and callbacks cannot launch or finalize the same run twice (#887).
• Restored packaged desktop agent controls after password login by granting the verified loopback owner session the narrow local-agent:run capability while keeping production localhost bypass disabled (#887).
• Made packaged desktop port selection detect listeners on both IPv4 and IPv6 loopback and kept dev server/web fallback ports distinct, preventing Electron from attaching to an unrelated local server or colliding with itself.
• Kept mobile notifications above the safe-area-aware bottom navigation so long-lived toasts cannot intercept touch input after Task Detail closes (#869).

5.2.4 - 2026-07-13

Fixed

• Restored the native macOS Edit menu so Cut, Copy, Paste, Select All, and their standard keyboard shortcuts work in desktop text and password fields (#842).
• Rebuilt the password-recovery actions as one full-width vertical stack with consistent spacing, and added password-manager autocomplete metadata to the login and recovery password fields (#842).

5.2.3 - 2026-07-13

Changed

• Enforced the tracked runtime security artifact guard in pre-commit and CI security gates, with documented rotation and private-evidence handling requirements (#836).

Fixed

• Rebuilt the desktop login actions as a stable vertical Mantine stack with a full-width primary action, centered recovery action, consistent spacing, and regression coverage for the packaged layout (#839).

5.2.2 - 2026-07-12

Added

• Added append-only JSONL activity storage with recovery and compatibility coverage for durable event history (#782, #808).

Changed

• Added canonical repository guidance for Codex, OpenClaw, and Hermes and documented validated provider behavior (#803).
• Routed web API requests through the shared authenticated client and retired the deprecated polling hook (#802).
• Updated the Codex SDK integration to 0.144.1 and pinned patched transitive development-tool dependencies (#792, #795).
• Added provider and command fields to WorkflowAgent in @veritas-kanban/shared to align with the server-side definition and prevent silent contract drift (#786).
• Added max_reroutes to FailurePolicy and retryRouteCount to WorkflowRun in both the shared and server workflow type contracts (#780, #786).

Fixed

• Hardened file-storage mutation ordering, rollback behavior, and integrity recovery across concurrent writes (#804).
• Fixed runtime lifecycle leaks, async registry persistence, attempt reconciliation, and cross-platform path helpers (#774, #779, #781, #783, #801).
• Normalized security artifact paths before matching so mixed-case paths cannot bypass expected classification (#807).
• Reworked Scoring Profiles into a compact list/detail flow at phone widths, preserved the selected profile across create and duplicate actions, and guarded unsaved drafts when returning to the list, changing tabs, or leaving the scoring surface (#813).
• Replaced broad and layout-property dashboard transitions with instant chart updates, explicit short hover feedback, and accessible immediate section expansion, including intentional reduced-motion behavior (#814).
• Made SQLite activity ordering deterministic for multiple events written in the same millisecond, preserving the newest-first feed contract and stable retention trimming (#823).
• Removed untracked asynchronous directory creation from AttachmentService construction; attachment and archive paths are now created lazily by the operations that need them (#825).
• Prevented compact bottom-navigation labels from overlapping, kept the mobile header within 320 px, and restored a fixed, safe-area-aware Board Chat entry point (#811).
• Kept mobile Settings navigation visible and touch-sized in the dialog content flow, and stacked the Product Mode summary and selector at compact widths (#810).
• Made dialog, sheet, alert, and Task Detail overlays honor reduced-transparency and increased-contrast preferences, strengthened modal focus boundaries, and dismissed task-card tooltips when opening Task Detail (#815).
• Restored keyboard drag-and-drop parity on the Kanban board with spatial cross-column movement, empty-column targeting, same-column reordering, accurate screen-reader announcements, mutation rollback, and post-move focus restoration (#812).
• Kept Electron's runtime API external in Vite 8/Rolldown desktop builds, preventing the npm executable-path shim from replacing app, BrowserWindow, contextBridge, and related native APIs in emitted main and preload artifacts; desktop builds now fail closed if the shim reappears (#809).
• Isolated QMD result normalization coverage from unrelated persistent search collections and restored test environment state only after temporary search roots are removed, eliminating the intermittent teardown race (#793).
• Human-gate blocking (on_false.escalate_to: human) now transitions the run to blocked instead of failed via the new HumanGateBlockError typed exception; approveGateStep and rejectGateStep service methods replace the unimplemented route stubs that previously discarded state (#778).
• retry_step cross-step rerouts are now bounded by a persisted retryRouteCount field; the default cap is 10 and is configurable via on_fail.max_reroutes; exhaustion respects on_exhausted policy or fails deterministically (#780).
• WorkflowRunService private NotFoundError and ValidationError classes replaced with the shared AppError hierarchy so workflow domain errors map to stable 404/400 HTTP responses instead of HTTP 500 (#785).
• BlockingService now enforces dependencies.depends_on (canonical dependency model) in addition to legacy blockedBy; the tasks route transition guard checks both fields so depends_on relationships block in-progress transitions (#787).

5.2.1 - 2026-06-29

Fixed

• Blocked disabled communication adapters from ingesting inbound human replies into Squad Chat (#760, #764).
• Made delegated workspace intake recoverable when config persistence fails after target task creation (#761, #764).
• Verified queue monitor candidate selection already prioritizes runnable work before the bounded candidate cap (#762).

Changed

• Refreshed the static docs homepage to remove brittle launch-era star and test count claims from current-release copy (#763, #764).
• Bumped the source metadata from 5.2.0 to 5.2.1 across package manifests and operator-facing version docs (#765).
• Stabilized workflow run metadata coverage and file-backed ChatService startup directory setup so release CI does not leak asynchronous test setup errors.

5.2.0 - 2026-06-26

Added

• Added workspace team roster manifests and capability routing rules for capability-aware delegation (#734).
• Added cross-workspace capability discovery and delegated work intake (#735).
• Added policy-gated queue intake monitoring for GitHub issues and agent runs (#736).
• Added a unified recurring work scheduler for monitors, workflows, and digests (#737).
• Added Squad Chat threads, unread state, mentions, and search (#738).
• Added bidirectional human reply adapters for Squad Chat and notifications (#739).
• Added ceremony enforcement for design reviews and failure retrospectives (#740).
• Added reflection-to-memory promotion for agent corrections and lessons (#741).
• Added external tracker schema introspection for configurable work items (#742).

Changed

• Bumped the release from 5.1.0 to 5.2.0 across the root, shared, server, web, CLI, MCP, and desktop package manifests.
• Updated the README version badge, API reference, v5 release notes, and MCP documentation footer for 5.2.0.

Fixed

• Fixed CLI/MCP compatibility smoke checks so missing VK_API_KEY records explicit skip results after local package/build validation instead of failing unrelated release gates (#753).
• Removed vulnerable frontmatter dependencies from the server by using the existing YAML parser and preserving malformed-prompt fallback behavior (#754).
• Upgraded DOMPurify and kept the Mantine QA gate below the initial bundle budget by deferring drag/drop code outside the first page load (#754, #755).

5.1.0 - 2026-06-18

Added

• Added shared live run sessions with workspace-scoped view, co-drive, and fork permissions. Shared viewers receive live output and run events, editors send attributed messages and mobile-safe approval responses, and forks create linked tasks without mutating parent run state (#721, #731).
• Added reusable agent profile packages with YAML/JSON validation, import, export, metadata editing, and profile-aware agent launch posture (#720).
• Added enforceable agent run budgets for token, cost, tool-call, runtime, retry, and fan-out limits with auditable enforcement decisions (#722).
• Added sandbox policy presets for filesystem, network, environment, and credential access with dry-run validation before launch (#723).
• Added multi-participant decision review sessions with independent responses, critique rounds, synthesis packets, work-product attachment, and decision audit links (#724).

Changed

• Bumped the release from 5.0.1 to 5.1.0 across the root, shared, server, web, CLI, MCP, and desktop package manifests.
• Updated the README version badge, API reference, release notes, permission coverage manifest, and MCP documentation footer for 5.1.0.

Fixed

• Fixed Docker source builds by restoring required Dockerfile inputs to the build context (#725).

5.0.1 - 2026-06-12

Changed

• Bumped the release from 5.0.0 to 5.0.1 across the root, shared, server, web, CLI, MCP, and desktop package manifests.
• Updated the README version badge and MCP documentation footer for 5.0.1.

Fixed

• Made MCP write tools return concise confirmations instead of echoing full task JSON payloads with complete comment history, reducing token usage for agent-heavy boards (#714, #715). Thanks to @Cob-AI for the report.
• Stabilized the Mantine task detail Progress tab test under CI load.
• Fixed the local macOS desktop packaging smoke check to follow the configured app bundle name.

5.0.0 - 2026-06-05

Added

• Added v5 release readiness docs covering upgrade/install/admin paths, compatibility matrix, release channels, rollback limits, release notes, and deferred post-GA platform backlog.

Changed

• Bumped the release from 4.3.2 to 5.0.0 across the root, shared, server, web, CLI, MCP, and desktop package manifests.
• Updated the README version badge and MCP documentation footer for 5.0.0.
• Expanded release validation to cover the desktop package version and required v5 release documentation.

[4.3.2] - 2026-05-16

Added

• Added pnpm validate:release to verify workspace version alignment, required release files, the README version badge, changelog headings, built artifacts, optional GitHub tag/release state, and optional production Docker image builds.
• Added .github/workflows/scheduled-qa.yml so Playwright and k6 smoke/load checks can run on a scheduled cadence outside the fast pull request CI path.
• Added the full codebase audit report at docs/CODEBASE-AUDIT-2026-05-16.md, including completed remediation notes and linked follow-up issues for the remaining larger work.
• Added dedicated URL validation regression coverage for blocked localhost, private-network, and unsafe redirect webhook targets.

Changed

• Bumped the release from 4.3.1 to 4.3.2 across the root, shared, server, web, CLI, and MCP package manifests.
• Updated the README version badge and MCP documentation footer for 4.3.2.
• Expanded the root build script and CI artifact checks to include the CLI and MCP packages so release validation covers every published/runtime surface.
• Split heavy web panels, chat, board, command palette, and task-detail surfaces out of the initial Vite bundle, removing oversized chunk warnings from the production build.
• Centralized web view metadata and task-detail tab metadata to reduce navigation drift, feature-gate drift, and duplicated labels.
• Added a lint warning budget gate after reducing current warning debt from 728 to 714, making future lint debt increases visible in CI.
• Stabilized full-suite test execution by tightening async Codex provider assertions, isolating Docker path filesystem mocks, and giving slower jsdom/route-integration tests realistic timeouts.

Fixed

• Hardened outbound webhook URL handling with shared validation, DNS/IP block checks, and redirect validation before webhook delivery.
• Fixed board drag reordering persistence, dependency-blocked task transitions, checkpoint clearing semantics, CLI/MCP version reporting, API envelope handling, and setup guidance found during the audit.
• Fixed Docker workspace dependency stages to include CLI and MCP package manifests before install/build steps.
• Fixed failure-alert, hook, policy, squad webhook, and transition hook services to use the hardened URL validation path consistently.

[4.3.1] - 2026-05-11

Security

• Restricted policy create, update, and delete routes to admin role only (#357).
• Restricted tool policy create, update, delete, and validation routes to admin role only (#355).
• Restricted agent permission level changes, permission field updates, and approval review to admin authority (#356).
• Restricted task-to-agent routing configuration updates to admin role only (#354).

Changed

• Updated workspace package versions, README badge, and MCP README footer from 4.3.0 to 4.3.1.

[4.3.0] - 2026-05-09

Added

• Veritas Cutover operating guide covering the authority model, HermesAgent roster, mandatory QA evidence gate, product/spec task template, research/revenue intake template, and Medik8 Cyprus-only task template.
• Agent Registry references for active and on-demand Veritas Cutover HermesAgent owners.
• Agent provider adapter contract for OpenClaw, Codex CLI, and Codex SDK start/stop/status/log/completion capabilities.
• Codex event mapping into activity, trace, token telemetry, and task deliverable surfaces.
• Mocked Codex JSONL fixture coverage for success, failure, malformed output, token usage, and file-change artifacts.

Changed

• Updated workspace package versions and README badge from 4.2.5 to 4.3.0.
• Updated Codex roadmap docs from v4.2 tracking language to v4.3 completion language.
• Bumped zod from v3 to v4.4.3 across the MCP, server, and web packages, including server schema compatibility updates.
• Bumped safe chore dependencies: @typescript-eslint/* to 8.59.2, express-rate-limit to 8.5.0, hono to 4.12.18, postcss to 8.5.14, and shadcn to 4.7.0.
• Added dependency audit overrides for patched fast-uri and ip-address transitive versions.
• Created the dependencies and github-actions GitHub labels expected by Dependabot.

Acknowledgements

• Thanks to @asperty567 for the Veritas Cutover issue set and operating direction.

[4.2.5] - 2026-05-06

Added

• Codex Settings health endpoint at GET /api/settings/codex/health.
• Server-side checks for Codex CLI installation, CLI version, codex login status, Codex SDK import availability, and Codex CLI/SDK/Cloud agent profile readiness.
• Settings Agents tab Codex Health panel with refresh, readiness badges, version/auth details, and actionable recommendations.
• Frontend API and hook support for Codex health checks.
• Route coverage for the Codex Settings health endpoint.

Changed

• Updated workspace package versions and README badge from 4.2.4 to 4.2.5.
• Documented richer Codex Settings health checks as implemented, completing the requested v4.2 patch train.

[4.2.4] - 2026-05-06

Added

• Codex review action for task branch diffs through POST /api/diff/:taskId/codex-review.
• Codex SDK review runner that executes in read-only sandbox mode, asks for structured JSON findings, and maps findings to Veritas review comments.
• Review summary/task comment persistence plus review decision metadata when review saving is enabled.
• Frontend API and React Query hook support for triggering Codex reviews.
• Route coverage for Codex review validation and service invocation.

Changed

• Updated workspace package versions and README badge from 4.2.3 to 4.2.4.
• Documented Codex review actions as implemented while richer Settings health checks remain next in the v4.2 patch train.

[4.2.3] - 2026-05-06

Added

• Workflow-engine Codex agent steps for workflow agents configured with Codex provider metadata or Codex agent naming.
• Codex SDK streaming execution inside workflow step runs, including worktree-aware execution, final response capture, and persisted step output files.
• Workflow session tracking for Codex thread IDs through run.context._sessions, enabling reuse-mode workflows to resume the same Codex thread.
• Unit coverage for Codex-backed workflow step execution.

Changed

• Updated workspace package versions and README badge from 4.2.2 to 4.2.3.
• Documented Workflow Codex steps as implemented while review actions and richer Settings checks remain next in the v4.2 patch train.

[4.2.2] - 2026-05-06

Added

• Codex Cloud delegation through GitHub issue, issue-comment, and PR-comment workflows.
• POST /api/github/codex/delegate for creating scoped @codex GitHub delegation prompts from Veritas tasks.
• Disabled-by-default codex-cloud agent profile and codex-cloud provider metadata support.
• Cloud delegation attempt metadata with cloudUrl and cloudTarget fields.
• Task comments that link successful Codex Cloud delegations back to the GitHub artifact.
• Frontend API hook for Codex Cloud delegation.

Changed

• Updated workspace package versions and README badge from 4.2.1 to 4.2.2.
• Documented Codex Cloud delegation as implemented while workflow steps, review actions, and richer Settings checks remain next in the v4.2 patch train.

[4.2.1] - 2026-05-06

Added

• Codex SDK session provider powered by @openai/codex-sdk, enabling durable local Codex threads for code-task attempts.
• Disabled-by-default codex-sdk agent profile for teams that want SDK-backed sessions alongside the existing codex CLI profile.
• Codex SDK attempt logging for streamed thread, turn, item, error, and usage events.
• Codex SDK token telemetry mapping and stop support through abortable SDK turns.
• threadId attempt metadata so SDK session IDs are preserved on active and completed attempts for follow-up workflows.

Changed

• Updated workspace package versions and README badge from 4.2.0 to 4.2.1.
• Documented Codex SDK sessions as implemented while keeping Cloud delegation, workflow steps, review actions, and richer Settings checks in the v4.2 patch train.

[4.2.0] - 2026-05-06

Added

• OpenAI Codex integration roadmap for v4.2, covering local Codex CLI execution, Codex SDK sessions, Codex Cloud delegation, MCP setup, workflow-engine execution, telemetry, review automation, and release QA (#310)
• Codex Integration SOP with operational guidance for CLI, SDK, Cloud, MCP, telemetry mapping, reviews, workflow steps, escalation, and release validation (#310)
• Codex workflow examples for local feature work, Codex review, SDK follow-up sessions, workflow pipelines, Cloud delegation, MCP-first board maintenance, and release QA (#310)
• Codex-specific AGENTS.md guidance and MCP setup instructions for Veritas-aware Codex work (#310)
• Codex feature reference in docs/FEATURES.md, updated to distinguish implemented CLI support from planned SDK, Cloud, review, and workflow modes (#310)
• Built-in codex agent profile backed by local codex exec --sandbox workspace-write --json
• Local Codex CLI execution path for code-task attempts, including process launch in the task worktree, attempt logs, final summary capture, completion handling, stop support, and run/token telemetry mapping
• Config migration that appends missing built-in agents, including Codex, to existing configs without overwriting customized agents

Changed

• Updated workspace package versions and README badge from 4.1.0 to 4.2.0 (#311)

4.1.0 - 2026-05-04

Added

• QMD-backed search foundation with keyword fallback across active tasks, archived tasks, and docs (#290)
• In-app search dialog with backend and collection controls, fallback status, and task-result navigation (#291)
• Duplicate detection hints in the create-task dialog using active and archived task retrieval (#292)
• VERITAS chat context injection with compact, cited <veritas_context> blocks and per-message opt-out (#293)
• QMD index maintenance commands and API refresh endpoint:
  • pnpm qmd:setup
  • pnpm qmd:refresh
  • POST /api/search/index/refresh (#294)

Changed

• Documented the v4.1 QMD setup, search UI, duplicate detection, VERITAS context injection, and index maintenance workflows
• Updated package versions from 4.0.1 to 4.1.0

4.0.1 - 2026-05-04

Fixed

• Restored task detail side-popout spacing after the shadcn/ui v4 sheet padding change
• Prevented task detail tabs from being squeezed in code-task popouts by allowing horizontal tab overflow
• Mass archive failure caused by missing /api/tasks/bulk-archive-by-ids endpoint
• MIME validation test fixtures now use valid PNG IHDR chunks (#266)
• Status history tests isolated from real state
• Build order: shared package built before server and web (#244)

Changed

• Batch file reads use bounded concurrency for improved task loading performance (#255)
• Shared package exports governance, policy, and workflow types (#256)
• Bumped pnpm/action-setup from v4 to v5 (#260)

Security

• Required N8N_WEBHOOK_SECRET for unauthenticated n8n webhook requests
• Hardened n8n webhook secret comparison, attachment type checks, filename sanitization, and path containment
• Patched high and moderate dependency advisories for @xmldom/xmldom, hono, postcss, and sanitize-html
• Hardened localhost bypass and broadcast frontmatter parsing (#242)
• Startup init failures now fatal with shutdown timeouts (#241)
• Config cache stampede prevention and corrupted activity file logging (#240)
• WebSocket event listener leak fix and message rate limiting (#239)
• Promise handling fixes preventing crashes and data loss (#238)
• Enforced HS256 algorithm in JWT verification (#237)

Docs

• v4.0 governance security audit (#265)
• Comprehensive unit tests for v4.0 governance services (#263)
• Feedback, chat, and system-health route tests (#257)

4.0.0 - 2026-03-21

✨ Highlights

Veritas Kanban 4.0 is the agent governance release. This release adds a complete Agent Governance Layer — configurable policies, behavioral drift detection, decision audit trails, output evaluation, and user feedback analytics — alongside a fully customizable dashboard widget grid, a Prompt Template Registry with version control, and a real-time Global System Health Status Bar.

All shadcn/ui components have been upgraded to v4 with Tailwind v4 integration. This release also adds MCP comment CRUD tools and project management tools, wires lifecycle hooks to the notification service, and ships seven new SOP guides plus updated API documentation.

Added

• Agent Policy & Guard Engine (#178) — Configurable tool/action policies with guard rules

  • Define policies with allow/deny/require-approval guard rules per tool and action
  • Policy evaluation engine with configurable precedence (deny-first or allow-first)
  • REST API: GET/POST /api/policies, GET/PUT/DELETE /api/policies/:id, POST /api/policies/:id/evaluate
  • Per-agent and per-project policy scoping
  • Built-in audit log for every policy decision

• Decision Audit Trail with Assumption Tracking (#179) — Log agent decisions with assumptions and outcomes

  • Structured decision records: decision text, confidence score, supporting evidence, assumptions
  • Outcome tracking: record what happened after a decision and whether assumptions held
  • Full-text search and filtering by agent, task, confidence range
  • REST API: GET/POST /api/decisions, GET/PUT/DELETE /api/decisions/:id, POST /api/decisions/:id/outcome

• Agent Output Evaluation & Scoring Framework (#180) — Quality scoring for agent outputs

  • Scoring profiles with weighted criteria: RegexMatch, KeywordContains, NumericRange, CustomExpression
  • Composite scoring methods: weightedAvg, minimum, geometricMean
  • Evaluation history with per-scorer breakdown and per-dimension explanations
  • REST API: GET/POST /api/scoring/profiles, GET/PUT/DELETE /api/scoring/profiles/:id, POST /api/scoring/evaluate, GET /api/scoring/history

• Behavioral Drift Detection & Alerting (#181) — Detect and alert on agent behavior changes

  • Metric tracking with configurable baselines and alert thresholds
  • Drift status lifecycle: ok → warning → alert → resolved
  • Automatic timestamp of detection and resolution events
  • REST API: GET/POST /api/drift, GET/PUT/DELETE /api/drift/:id, POST /api/drift/:id/resolve

• User Feedback Loop with Sentiment Analytics (#182) — Collect and analyze user feedback on agent outputs

  • Feedback collection with sentiment (positive/neutral/negative) and category tagging
  • Aggregate analytics: sentiment breakdowns, trends over time, category distributions
  • REST API: GET/POST /api/feedback, GET/DELETE /api/feedback/:id, GET /api/feedback/analytics

• Draggable & Resizable Dashboard Widget Grid (#183) — Customizable dashboard layouts

  • Drag-and-drop widget repositioning with smooth animations
  • Resizable widgets with snap-to-grid behavior
  • Layout persistence via settings.json
  • Add/remove widgets from the widget library

• Prompt Template Registry with Version Control (#184) — Manage and version prompt templates

  • Template CRUD with variable extraction (e.g., {{agent_name}}, {{task_context}})
  • Full version history with changelog entries; rollback to any version
  • Usage tracking: model, tokens, rendered output (optional)
  • Preview rendering with sample variable injection
  • REST API: GET/POST /api/prompt-registry/templates, GET/PUT/DELETE /api/prompt-registry/templates/:id, GET /api/prompt-registry/templates/:id/versions, POST /api/prompt-registry/templates/:id/versions, GET /api/prompt-registry/templates/:id/versions/:versionId, POST /api/prompt-registry/templates/:id/preview, GET /api/prompt-registry/templates/:id/stats, POST /api/prompt-registry/templates/:id/usage

• Global System Health Status Bar (#185) — Real-time system health monitoring

  • Five health levels: stable → reviewing → drifting → elevated → alert
  • Three signal categories: system (storage/disk/memory), agents (online/offline counts), operations (success rate, recent runs)
  • Persistent header bar with expand/collapse for detail panel
  • REST API: GET /api/v1/system/health

• MCP comment CRUD tools (PR #206) — Full comment lifecycle from MCP

  • New MCP tools: add_comment, list_comments, get_comment, update_comment, delete_comment
  • Works with both task-level and sprint-level comments

• MCP project management tools (PR #227) — Full project lifecycle from MCP

  • New MCP tools: list_projects, get_project, create_project, update_project, delete_project, get_project_stats, reorder_projects

• Lifecycle hooks wired to notification service (PR #201) — Hook outcomes now trigger real-time notifications

  • Hooks with notify: true push events through the notification service
  • Integrates with @mention notification system for addressable hook alerts

• Comprehensive MCP server documentation at docs/mcp/README.md — architecture, quickstart, full tool catalog with examples, security model, troubleshooting playbook, and FAQ

• Condensed root README MCP section with link to dedicated docs

Changed

• Upgraded shadcn/ui components to v4 (#186, PR #219) — All UI components updated for shadcn/ui v4 compatibility with Tailwind v4 integration; breaking prop changes resolved across the component tree

Fixed

• Squad chat panel scroll (PR #225, #224) — Panel now properly scrolls to latest message; overflow clipping bug in nested flex containers resolved
• False cycle detection in dependency API (PR #208, #188) — DFS algorithm no longer flags valid multi-path dependencies as cycles when two nodes share a common ancestor
• TypeScript build errors (PR #207, #177) — Resolved type mismatches introduced by shadcn/ui v4 and Zod 4 type inference changes
• SystemHealthBar successRate percentage display (PR #212, #211) — successRate is now correctly formatted as a percentage (0–100) instead of a decimal (0–1)

3.3.3 - 2026-03-01

✨ Highlights

Veritas Kanban 3.3.3 is a patch correction release delivering stability, security, and performance fixes on top of 3.3.2, plus full Zod 4 API migration completing the TypeScript compilation fix.

Fixed

• #162 — Complete Zod 4 API migration — Resolved 50+ TypeScript compilation errors blocking CI/CD

  • ZodError.errors → ZodError.issues across all route files and middleware
  • z.record(valueSchema) → z.record(z.string(), valueSchema) for explicit key types
  • Schema defaults corrected from string to proper numeric/boolean types (PORT, VERITAS_AUTH_ENABLED, RATE_LIMIT_MAX)
  • required_error → message in number schemas (agents.ts, agent-schemas.ts)
  • Affected: validate.ts, agent-registry.ts, workflows.ts, lifecycle-hooks.ts, tasks.ts, reports.ts, env.ts

• #165 — SSRF protection for webhook URLs — Added server-side request forgery safeguards for all outbound webhook destinations (replaces #163)

Added

• Orchestrator Delegation Enforcement — Full enforcement gate for orchestrator delegation

  • Orchestrator agent selector in Settings → Enforcement (dropdown of enabled agents)
  • Active/Inactive status badge showing enforcement state
  • Warning banner when delegation enabled but no agent selected
  • Section auto-disables when delegation toggle is off
  • Zod-validated orchestratorAgent field (string, max 50 chars)
  • POST /api/agent/delegation-violation endpoint for violation reporting
  • Auto-posts violations to squad chat when squad chat enforcement is enabled

• Enforcement Gate Toast Notifications — Enhanced error feedback for all enforcement gates

  • Gate-specific titles and actionable guidance for REVIEW_GATE, CLOSING_COMMENTS_REQUIRED, DELIVERABLE_REQUIRED, ORCHESTRATOR_DELEGATION
  • 10-second toast duration for enforcement messages (up from 5s)
  • BulkActionsBar surfaces gate details on bulk move failures

• Dashboard Enforcement Indicator — At-a-glance enforcement status

  • Shows active/total gate count with color-coded shield icon (green/amber/gray)
  • Individual gate dots (green = active, gray = off)
  • Renders in dashboard status bar alongside refresh timestamp

Performance

• #167 — WebSocket broadcast batching — Batch broadcasts to prevent event loop blocking under high-frequency update load (replaces #164/#166)

---

[3.3.2] - 2026-03-01

✨ Highlights

Veritas Kanban 3.3.2 delivers sprint management for CLI + MCP surfaces alongside task↔agent state synchronization, security hardening of the sync auth boundary, and a circuit breaker test suite.

Added

• #161 — Sprint Management (CLI + MCP) — Full sprint CRUD and task integration from the command line and MCP

  • vk sprint list — list sprints with --hidden / --json flags
  • vk sprint create — create sprints with optional description
  • vk sprint update — update label, description, visibility
  • vk sprint delete — delete sprints (with --force for non-empty)
  • vk sprint close — archive completed tasks in a sprint
  • vk sprint suggestions — show sprints ready to archive
  • vk list -S <sprint> / vk create -S <sprint> / vk update -S <sprint> — task sprint integration
  • MCP tools: list_sprints, create_sprint, update_sprint, delete_sprint, close_sprint, sprint_suggestions
  • Sprint field surfaced on list_tasks, create_task, update_task MCP tools

• #155 — Task↔Agent State Sync + Reconciliation — Bi-directional sync engine keeping task state consistent with agent execution state; reconciliation pass resolves diverged records

• #156 — Circuit Breaker Test Suite — 18 new unit tests covering circuit breaker open/half-open/closed transitions, timeout behaviour, and error thresholds

Fixed

• #159 — Harden task-agent sync auth boundary — Tightened authentication checks on sync routes to prevent unauthorized state manipulation (#157 #158)

---

[3.3.1] - 2026-02-28

Fixed

• #153 — Express 5 wildcard route compatibility — Updated route patterns for Express 5.x compatibility
• Security: upgrade hono to >=4.12.2 — Addressed GHSA-xh87-mx6m-69f3 vulnerability

Changed

• #148 — Bump vite from 6.4.1 to 7.3.1 — Major version upgrade with performance improvements
• #149 — Bump tailwindcss from 3.4.19 to 4.2.1 — Major version upgrade with new features
• #154 — Bump production dependencies (8 updates) — Routine dependency maintenance

---

[3.3.0] - 2026-02-15

✨ Highlights

Veritas Kanban 3.3 delivers production-grade task intelligence — four features that significantly improve task management capabilities, reliability, and accessibility. This release represents the culmination of rigorous cross-model development: all features authored by Sonnet, reviewed by Codex, and scored 10/10 across security, reliability, performance, and accessibility through our 4x10 review gate.

Key improvements:

• Dependency graph prevents circular dependencies through DFS cycle detection that traverses bidirectional relationships
• Crash-recovery checkpointing protects credentials with auto-sanitization of 20+ secret patterns plus regex value detection
• Observational memory preserves critical context across agent sessions with full-text search and importance scoring
• Agent filtering enables precise task queries by agent name with proper input sanitization

Added

• #122 — Task Dependencies Graph

  • Bidirectional dependency model (depends_on / blocks) with cycle detection
  • DFS algorithm traverses both directions to prevent dependency loops
  • Dependency graph API with recursive tree traversal (GET /api/tasks/:id/dependencies)
  • UI: DependenciesSection component with add/remove for both directions
  • TaskCard badges showing dependency count
  • Zod validation on dependency routes
  • Batch-loaded graph traversal (eliminated N+1 queries)
  • Full keyboard + ARIA accessibility

• #123 — Crash-Recovery Checkpointing

  • Save/resume/clear API (POST/GET/DELETE /api/tasks/:id/checkpoint)
  • Auto-sanitization of secrets (20+ key patterns + regex value detection)
  • 1MB size limit on checkpoint state with 24h expiry and automatic cleanup
  • Resume counter tracks restart attempts
  • Sub-agent checkpoint context injection into prompts
  • Array sanitization (nested objects + primitive strings)
  • NaN timestamp handling
  • ARIA-accessible checkpoint UI in TaskCard + TaskDetailPanel

• #124 — Observational Memory

  • Add/view/delete observations per task (decision, blocker, insight, context types)
  • Importance scoring (1-10) with visual badges
  • Full-text search across all task observations (GET /api/observations/search)
  • Paginated search results (limit/offset, max 200)
  • Timeline view with type-colored badges
  • Activity logging for audit trail
  • XSS prevention via sanitizeCommentText()
  • ARIA-accessible range slider + decorative icon handling

• #125 — Agent Filter

  • GET /api/tasks?agent=name query parameter
  • Input sanitized (trim + 100 char cap)
  • Works with existing pagination and filters
  • JSDoc/OpenAPI documented

Technical Notes

All features passed comprehensive security review:

• Input sanitization on all user-facing fields
• XSS prevention in observational memory comments
• Secret detection and auto-sanitization in checkpoints
• Cycle detection prevents infinite loops in dependency traversal

[3.2.1] - 2026-02-12

Fixed

• reviewGate enforcement — Now only applies to code task types (code, bug, feature, automation, system) — research, content, and custom types no longer blocked
• Defensive settings access — Prevents crash on missing config sections across all 8 settings tabs (General, Enforcement, Time Tracking, Agents, Projects, Sprints, Templates, Types)
• Settings toggle persistence — Fixed deepMergeDefaults overwriting user changes; now uses proper patch merging to preserve toggle states
• SharedResources toggle persistence — SharedResources and DocFreshness enabled/disabled states now correctly saved to settings.json
• Squad chat visibility — System messages now render as gray message bubbles (previously invisible dividers)
• Time tracking telemetry — Fixed corrupt 17K-hour entry, added 7-day (604,800,000 ms) cap on durationMs validation
• Archived task reappearance — Tasks no longer reappear on board after archival (orphaned files from title changes now cleaned up via findTaskFile)
• EnforcementTab formatting — reviewGate warning text properly formatted, "Quality Gates" duplicate header removed

Added

• PRD-Driven Autonomous Development guide — Complete 961-line guide at docs/features/prd-driven-development.md covering setup, agent execution workflow, OAuth2 example, configuration tips, and troubleshooting
• Time tracking bugfix post-mortem — Documented 17K-hour telemetry bug in docs/bugfix-time-tracking-17k-hours.md

Changed

• Consistent gray styling — Squad chat system messages now use muted theme tokens for subtle gray appearance
• Theme token usage — Settings UI updated to use Shadcn theme tokens instead of hardcoded Tailwind colors

[3.2.0] - 2026-02-11

Added

• Markdown Editor — Rich markdown editing for task descriptions and comments with formatting toolbar, live preview, keyboard shortcuts (Ctrl+B/I/K), syntax highlighting, and dark mode support. Configurable via Settings → Tasks.
• Shared Resources Registry — Define reusable resources (prompts, guidelines, skills, configs, templates) and mount them across projects. Full CRUD API with mount/unmount endpoints. Configurable via Settings → Shared Resources.
• Documentation Freshness Tracking — Track document staleness with freshness scores, alerts, and optional auto-review task creation. API at /api/doc-freshness. Configurable via Settings → Doc Freshness.

Fixed

• Persist runtime auth/config state to the Docker volume by routing .veritas-kanban paths through getRuntimeDir() and migrating legacy files on startup (security.json, agent registry, lifecycle hooks, error analyses, agent permissions).
• Added Docker migration guidance for recovering auth state after rebuilding containers.
• Dark mode Lessons Learned display bug fixed (text was unreadable in dark mode)
• Plain text card previews for task descriptions (markdown rendering disabled on board cards for performance)
• JWT rotation test flakiness fixed (timing issue resolved)

[3.1.0] - 2026-02-10

Added

• 6 structural enforcement gates (all disabled by default):
  • squadChat, reviewGate, closingComments, autoTelemetry, autoTimeTracking, orchestratorDelegation
• Comprehensive enforcement documentation with a dedicated "For AI Agents" section
• Agent SOP enforcement awareness updates

Fixed

• Enforcement gate logic (correct === true checks)
• TelemetryService supports backdated timestamps
• TelemetryService respects DATA_DIR in Docker environments
• Zod schema allows optional timestamp on telemetry events
• Template YAML serialization (recursive cleanForYaml)
• Case-insensitive agent color lookup in squad chat

Issue: #115

---

[3.0.0] - 2026-02-09

✨ Highlights

Veritas Kanban 3.0 ships the workflow engine — a deterministic multi-step agent orchestration system that transforms VK from an ad-hoc task board into a repeatable, observable, and reliable agent execution platform. Think GitHub Actions for AI agents.

14,079 lines of code shipped across 6 major phases:

• Phase 1: Core workflow engine (~7,091 lines)
• Phase 2: Run state management (~1,409 lines)
• Phase 3: Frontend + real-time updates (~3,069 lines)
• Phase 4: Advanced orchestration (~2,255 lines)
• Dashboard: Monitoring & health metrics (~2,050 lines)
• Policies & Sessions: Tool policies + session isolation (~1,200 lines)

Added

Core Workflow Engine (Phase 1 — #107)

• YAML workflow definitions — Define multi-step agent pipelines as version-controlled YAML files
• Workflow CRUD API — 9 REST endpoints for workflow management:
  • GET /api/workflows — List all workflows
  • GET /api/workflows/:id — Get workflow definition
  • POST /api/workflows — Create workflow
  • PUT /api/workflows/:id — Update workflow (auto-increment version)
  • DELETE /api/workflows/:id — Delete workflow
  • POST /api/workflows/:id/runs — Start a run
  • GET /api/workflow-runs — List runs (filterable by workflow, task, status)
  • GET /api/workflow-runs/:id — Get run details
  • POST /api/workflow-runs/:id/resume — Resume blocked run
• Sequential step execution — Execute workflow steps in order with retry routing
• Step types: agent — Agent steps spawn OpenClaw sessions with prompts (Phase 1 placeholder)
• Template rendering — Basic {{variable}} and {{nested.path}} substitution in step inputs
• Acceptance criteria validation — Simple substring matching for step completion checks
• Retry routing — Three strategies: retry same step, retry different step (retry_step), escalate to human
• Workflow snapshot — YAML saved in run directory for version immutability
• RBAC — Role-based access control with ACL files (.acl.json)
• Audit logging — All workflow changes logged to .audit.jsonl
• Zod validation — Schema validation for workflow definitions
• Storage structure — File-based persistence in .veritas-kanban/workflows/ and .veritas-kanban/workflow-runs/

Run State Management (Phase 2 — #113, #110, #111, #108)

• Persistent run state — Run state survives server restarts via checkpoint timestamps
• Progress file tracking (#108) — Shared progress.md per run for context passing between steps
• Step output resolution — Template variables like {{steps.plan.output}} resolve from previous step outputs
• Retry delays — Configurable retry_delay_ms to prevent rapid retry loops
• Tool policies (#110) — Role-based tool restrictions:
  • 5 default roles: planner, developer, reviewer, tester, deployer
  • Each role has allowed/denied tool lists
  • Custom role CRUD via /api/tool-policies endpoints
  • Tool filter passed to OpenClaw sessions (ready for Phase 3 integration)
• Fresh session per step (#111) — Each workflow step can spawn a fresh OpenClaw session:
  • Session modes: fresh (new session) or reuse (continue existing)
  • Context injection: minimal, full, or custom (specify which steps to include)
  • Cleanup modes: delete (terminate after step) or keep (leave running for debugging)
  • Configurable timeout per step
• Session tracking — Session keys stored in run.context._sessions per agent
• Backward compatible — Legacy fresh_session: false maps to session: reuse

Frontend + Real-Time Updates (Phase 3 — #107 frontend)

• WorkflowsPage — Browse and start workflow runs:
  • Grid view of all workflows with metadata (name, version, agents, steps)
  • "Start Run" button → calls POST /api/workflows/:id/run
  • Active run count badges per workflow
  • Search filter
  • Lazy-loaded in App.tsx
• WorkflowRunView — Live step-by-step workflow run visualization:
  • Real-time step progress display
  • Each step shows status, agent, duration, retry count, output preview, errors
  • Color-coded step status (green=completed, blue=running, red=failed, yellow=skipped, gray=pending)
  • "Resume" button for blocked runs
  • Overall progress bar (step X of Y)
  • Auto-updates via WebSocket workflow:status events
• WorkflowRunList — Filter and browse runs:
  • Filter by status (all, running, completed, failed, blocked, pending)
  • Click to open WorkflowRunView
  • Progress bars, duration tracking
• WorkflowSection — Run workflows from TaskDetailPanel:
  • Shows available workflows
  • Displays active runs for current task
  • "Start" button with task context
  • Dialog modal
• Navigation tab — "Workflows" tab added to header with icon
• WebSocket integration — Real-time updates via workflow:status events
• Polling fallback — Aggressive polling (10-30s) when WebSocket disconnected, safety-net polling (120s) when connected
• ~75% reduction in API calls when WebSocket connected

Advanced Orchestration (Phase 4 — #112, #113)

• Loop steps (#112) — Iterate over collections with progress tracking:
  • Configuration: over, item_var, index_var, completion policy, fresh_session_per_iteration
  • Completion policies: all_done, any_done, first_success
  • Loop state tracking: totalIterations, currentIteration, completedIterations, failedIterations
  • Output per iteration saved to step-outputs/<step-id>-<iteration>.md
  • Loop variables accessible in templates: {{loop.index}}, {{loop.total}}, {{loop.completed}}
  • Max 1000 iterations safety limit
  • continue_on_error flag to skip failed iterations
  • verify_each and verify_step for post-iteration validation (wired in types, executor pending)
• Gate steps — Conditional blocking with approval workflow:
  • Boolean expressions: {{test.status == "passed" and verify.decision == "approved"}}
  • Supports ==, and, or operators with variable access
  • Blocking behavior: run status changes to blocked if condition fails
  • Approval API: POST /api/workflow-runs/:runId/steps/:stepId/approve and /reject
  • Escalation policies: escalate_to: human blocks, timeout support
• Parallel steps — Fan-out/fan-in execution:
  • Execute multiple sub-steps concurrently via Promise.allSettled()
  • Completion criteria: all (all must succeed), any (at least one), N (at least N sub-steps)
  • fail_fast flag aborts remaining sub-steps on first failure
  • Aggregated JSON output with per-sub-step status, outputs, errors
  • Max 50 concurrent sub-steps (soft limit)
• Enhanced acceptance criteria — Regex patterns and JSON path equality:
  • Regex: /^STATUS:\s*done$/i
  • JSON path: output.decision == "approved"
  • Backward compatible substring matching
• Expression evaluator — Safe variable access and boolean logic (no arbitrary code execution)

Workflow Dashboard (#114)

• Summary cards (6 metrics):
  • Total workflows defined
  • Active runs (currently executing)
  • Completed runs (period-filtered: 24h/7d/30d)
  • Failed runs (period-filtered)
  • Average run duration
  • Success rate (%)
• Active runs table — Live-updating list of currently executing runs:
  • Workflow ID, status badge, started time, duration, current step, progress (step X/Y)
  • Click to open WorkflowRunView
  • Real-time updates via WebSocket
  • Visual progress bars
• Recent runs history — Last 50 workflow runs:
  • Sortable by status (all/completed/failed/blocked/pending)
  • Run ID, status badge, start time, duration, steps completed
  • Click to open WorkflowRunView
• Workflow health metrics — Per-workflow stats:
  • Success rate
  • Average duration
  • Run counts (total, completed, failed)
  • Visual health indicators (green/yellow/red based on success rate)
• Real-time updates — WebSocket-driven with polling fallback (30s when disconnected, 120s when connected)
• Backend endpoints:
  • GET /api/workflow-runs/active — Currently running workflows
  • GET /api/workflow-runs/stats?period=7d — Aggregated statistics (total workflows, active, completed, failed, avg duration, success rate, per-workflow breakdown)
• Navigation — "Dashboard" button in WorkflowsPage header

Changed

WebSocket Refactor (Phase 3)

• All hooks now WebSocket-primary — Polling is safety net only
• Connected behavior — 120s polling intervals (safety net)
• Disconnected behavior — Aggressive polling resumes (10-30s intervals)
• Events — task:changed, agent:status, telemetry:event, workflow:status
• Broadcast service — Centralized broadcastWorkflowStatus() function sends full run state (no extra HTTP fetches needed)
• 13 hooks/components refactored — All polling intervals updated

Security

• ReDoS protection — Regex patterns validated with size/complexity limits
• Expression injection prevention — Template evaluator only supports safe variable access and boolean operators
• Parallel DoS limits — Max 50 concurrent sub-steps in parallel execution
• Gate approval validation — Authentication and permission checks on approval endpoints
• Path traversal protection — sanitizeFilename on all file writes

Performance

• ~75% reduction in API calls — When WebSocket connected, polling drops to 120s safety-net intervals
• Progress file size cap — 10MB limit prevents unbounded growth
• Lazy-loaded frontend — WorkflowsPage, WorkflowDashboard only render when navigated to
• Memoized filters — useMemo for filtered workflows/runs
• Skeleton loading states — Shimmer placeholders during data fetch

---

[2.1.4] - 2026-02-09

Fixed

• Status Counter Accuracy (#104) — Sidebar task counts now use a dedicated GET /api/tasks/counts endpoint that returns total counts by status, independent of time-range filters applied to the board view

  • New: server/src/routes/tasks.ts — /api/tasks/counts endpoint returns { todo, in_progress, review, done, blocked, cancelled }
  • New: web/src/hooks/useTaskCounts.ts — dedicated React hook for sidebar counts
  • Updated: BoardSidebar.tsx — uses useTaskCounts() instead of deriving from filtered task list
  • Cache invalidation wired to task mutations so counts stay in sync

• Bulk Operation Timeouts (#105) — Bulk archive, status update, and backlog demote operations now use single API calls instead of N individual requests

  • New: POST /api/tasks/bulk-update — update status for multiple tasks in one call
  • New: POST /api/tasks/bulk-archive-by-ids — archive multiple tasks by ID array
  • New: POST /api/backlog/bulk-demote — demote multiple tasks to backlog in one call
  • All bulk endpoints validate array size (max 100) to prevent abuse
  • Operations run in parallel via Promise.allSettled() (~26× faster for large batches)
  • Updated: BulkActionsBar.tsx — rewired to use bulk endpoints instead of sequential loops
  • Updated: useTasks.ts, useBacklog.ts — new mutation hooks for bulk operations

Changed

• Squad Chat Documentation (#106) — Updated SQUAD_CHAT_IMPLEMENTATION.md to clarify that the model field is a structural JSON field in the API, not a text instruction

[2.1.3] - 2026-02-07

Fixed

• Docker Path Standardization (#102) — Comprehensive refactor: created shared paths.ts utility and migrated all 7 services to use it
  • New: server/src/utils/paths.ts — single source of truth for all path resolution
  • Refactored: task-service, activity-service, chat-service, audit-service, metrics/helpers, backlog-repository (7 files total)
  • Resolution priority: DATA_DIR / VERITAS_DATA_DIR env var → project root auto-discovery → fallback
  • Safety: Filesystem root guard prevents silent / resolution (the original EACCES bug)
  • Backwards compatible: Existing DATA_DIR configurations continue to work unchanged
  • Cross-model reviewed: 10/10/10/10 (GPT-5.1 authored, Claude Sonnet 4.5 reviewed)

[2.1.2] - 2026-02-07

Fixed

• Docker Path Resolution (#102) — Fixed WORKDIR resolution so services correctly find .veritas-kanban in containerized deployments
  • Root cause: Services use process.cwd()/.. to locate project root; with WORKDIR /app this resolved to / (filesystem root), causing EACCES: permission denied on container startup
  • Fix: Changed production WORKDIR to /app/server, ensured /app/tasks and /app/server are writable
  • Impact: Resolves permission denied errors when starting VK in Docker containers
  • Related: Issue #102 (Docker: Standardize .veritas-kanban path resolution across services)

[2.1.1] - 2026-02-07

Fixed

• Reverse Proxy Support (#100) — Added TRUST_PROXY environment variable for deployments behind nginx, Caddy, Traefik, Synology DSM, and other reverse proxies. Fixes express-rate-limit ValidationError and WebSocket authentication loops caused by untrusted X-Forwarded-For headers.
  • Supports hop counts (TRUST_PROXY=1), named values (loopback, linklocal), and subnet strings
  • TRUST_PROXY=true is blocked by default (security hardening — logs warning, falls back to no trust)
  • Disabled by default — no behavior change for existing deployments
  • Documentation added for nginx, Caddy, and Docker Compose configurations

Security

• Blocked TRUST_PROXY=true to prevent accidental trust-all-proxies misconfiguration on public-facing deployments

[2.0.0] - 2026-02-06

✨ Highlights

Veritas Kanban 2.0 is the multi-agent release. 18 features shipped across agent orchestration, dashboard analytics, lifecycle automation, and developer experience. This release transforms VK from a single-agent task board into a full multi-agent orchestration platform.

Added

Multi-Agent System (#28, #29, #30, #31)

• Agent Registry (#52) — Service discovery with heartbeat tracking, capabilities, live status, REST API for register/deregister/heartbeat/stats
• Multi-Agent Dashboard Sidebar (#28) — Real-time agent status cards in board sidebar, expandable details, color-coded status indicators (green=working, purple=sub-agent, gray=idle, red=error)
• Multi-Agent Task Assignment (#29) — Assign multiple agents to a single task, color-coded agent chips in task detail and board cards, shared helper utilities
• @Mention Notifications (#30) — @agent-name parsing in comments, thread subscriptions, delivery tracking, notification bell
• Agent Permission Levels (#31) — Intern / Specialist / Lead tiers with configurable approval workflows and autonomy boundaries

Dashboard Analytics (#57, #58, #59, #60, #61)

• Where Time Went (#57) — Time breakdown by project via telemetry data with color-coded project bars
• Activity Clock (#58) — 24-hour donut chart showing agent work distribution, sourced from status-history transitions
• Hourly Activity Chart (#59) — Bar chart with per-hour event counts, sourced from status-history
• Wall Time Toggle (#60) — Total Agent Time + Avg Run Duration with explanatory tooltips
• Session Metrics (#61) — Session count, success rate, completed/failed/abandoned tracking

Lifecycle & Automation

• Task Lifecycle Hooks (#72) — 7 built-in hooks (subtask-gate, assignee-required, blocked-reason, done-checklist, auto-archive, time-tracking, notification), 8 lifecycle events, custom hooks API
• Documentation Freshness (#74) — Steward workflow with freshness headers (fresh-days, owner, last-verified), 3-phase automation plan
• Error Learning Workflow (#91) — Structured failure analysis, similarity search for recurring issues, stats API. Inspired by @nateherk's Klouse dashboard concept.

Developer Experience

• Markdown Rendering (#63) — MarkdownText component for rich text in task descriptions and comments
• Cost Prediction (#54) — Multi-factor cost estimation model (tokens, compute, overhead) for task budgeting
• CLI Usage Reporting (#50) — vk usage command for token and cost reporting from the terminal
• Dashboard Widget Toggles (#92) — Show/hide individual dashboard widgets with settings gear and localStorage persistence
• Production Binding (#55) — VK_HOST and VK_PORT environment variables for flexible deployment
• Custom favicon — Purple scales-of-justice SVG icon replacing the default Vite favicon

Changed

• Timezone-aware metrics — Server reports its timezone dynamically in all API response meta; clients can request metrics in their local timezone via ?tz=<offset> query parameter
• Activity data source — Activity Clock and Hourly Activity Chart now pull from status-history (reliable state transitions) instead of activity.json
• Cost-per-task clickability — Enhanced hover states, border effects, and arrow indicator
• Archive optimistic updates — Archive mutations now remove tasks from cache immediately via onMutate, with rollback on error
• Agent naming convention — Agent names use ALL CAPS for acronyms (VERITAS, TARS, CASE, K-2SO, R2-D2, MAX)

Fixed

• Daily Activity 100% bug — Utilization was calculated using UTC dates but displayed in local timezone, causing incorrect percentages
• Feb 3 telemetry outlier — 66-minute run normalized to 19min (p95 level)
• Feb 2 telemetry outliers — 3 runs (15-19min range) normalized to 10min
• Registry stats interface mismatch — Frontend expected totalAgents/onlineAgents but server sent total/online; interface updated to match server

Security

• MCP SDK vulnerability patched — Updated @modelcontextprotocol/sdk from 1.25.3 to ^1.26.0 (GHSA-345p-7cg4-v4c7, cross-client data leak)
• Rate limiting documented — README now warns that VK does not include built-in rate limiting; reverse proxy recommended for public deployments

Maintenance

• 21 stale feature branches cleaned — Down to main only
• README roadmap updated — Reflects v2.0 shipped features, v1.6.0 and earlier history preserved
• Version bumped across all packages (root, server, web, shared, mcp)

Credits

• @nateherk — Error learning workflow inspired by Klouse dashboard concept
• @mvoutov — Documentation freshness inspired by BoardKit Orchestrator

---

[1.6.0] - 2026-02-05

✨ Highlights

• Activity Page Redesign — Streamlined to focus on status history with full-width layout, clickable task navigation, and color-coded status badges
• Task Templates UI (#39) — Full management interface for creating, editing, and instantiating task templates with blueprints
• Analytics API (#43) — New endpoints for timeline visualization and aggregate metrics (parallelism, throughput, lead time)
• Status Transition Hooks — Quality gates and automated actions for task status changes
• 7 GitHub Issues Closed — #47, #48, #49, #51, #53, #56, #82 verified complete and documented

Added

Activity Page Improvements

Full-Width Status History:

• Removed activity feed column — status history now spans full width
• Removed redundant "Status History" header label
• Daily summary panel retained above status history
• Cleaner, more focused interface for monitoring agent activity

Clickable Task Navigation:

• Status history entries now clickable to open task detail panel
• Keyboard accessible (Enter/Space to activate)
• Hover state indicates interactivity

Color-Coded Status Badges:

• Agent status colors:
  • working / thinking — Green
  • sub-agent — Purple
  • idle — Gray
  • error — Red
• Task status colors (Kanban columns):
  • todo — Slate
  • in-progress — Amber
  • blocked — Red
  • done — Blue
• Task titles colored to match their new status
• Uniform badge width for visual consistency

Task Status Changes:

• Now shows both agent status changes AND task status changes
• Task status changes display with kanban column colors
• Unified timeline view of all activity

Files:

• web/src/components/activity/ActivityFeed.tsx — Redesigned component
• web/src/hooks/useStatusHistory.ts — Updated color functions

Task Templates UI (#39)

Full management interface for task templates:

Templates Page (/templates):

• Grid view of all templates with category grouping
• Search and filter by category
• Quick actions: Edit, Preview, Delete, Create Task
• Empty state with helpful onboarding

Template Editor Dialog:

• Create and edit templates
• Configure task defaults (type, priority, project, agent)
• Add subtask templates with ordering
• Blueprint support for multi-task workflows
• Validation and error handling

Template Preview Panel:

• Read-only preview of template configuration
• Shows all defaults and subtasks
• One-click task creation from template

Files:

• web/src/components/templates/TemplatesPage.tsx
• web/src/components/templates/TemplateEditorDialog.tsx
• web/src/components/templates/TemplatePreviewPanel.tsx
• server/src/routes/templates.ts
• server/src/services/template-service.ts

Analytics API (#43)

New endpoints for advanced metrics and visualization:

Timeline Endpoint: GET /api/analytics/timeline

• Returns task execution timeline data
• Includes start/end times from time tracking
• Task assignments and status history
• Parallelism snapshots (concurrent tasks over time)
• Query params: from, to, agent, project, sprint

Metrics Endpoint: GET /api/analytics/metrics

• Aggregate metrics for a time period:
  • Parallelism factor (average concurrent tasks)
  • Throughput (tasks completed per period)
  • Lead time (creation to completion)
  • Agent utilization (working time per agent)
  • Efficiency metrics (tracked vs total time)
• Query params: sprint, from, to, project

Files:

• server/src/routes/analytics.ts
• server/src/services/analytics-service.ts
• server/src/schemas/analytics-schemas.ts
• docs/API-analytics.md — Swagger-style documentation

Status Transition Hooks

Quality gates and automated actions for task status changes:

Pre-Transition Gates:

• Must pass before status change is allowed
• Examples: require description, require time logged, require code review

Post-Transition Actions:

• Fire after status change succeeds
• Examples: notify channel, update external system, trigger automation

Configuration:

• Stored in .veritas-kanban/transition-hooks.json
• Configurable per-transition (e.g., in-progress → done)
• Enable/disable globally or per-rule

Files:

• server/src/services/transition-hooks-service.ts
• server/src/routes/transition-hooks.ts
• shared/src/types/transition-hooks.ts

CLI Setup Wizard

Interactive onboarding for new users:

vk setup

• Guided configuration of API URL and auth
• Creates .veritas-kanban/config.json
• Tests connection and validates setup
• Shows next steps and helpful commands

Files:

• cli/src/commands/setup.ts

Prompt Registry

10 copy/paste prompt templates in docs/prompt-registry/:

1. task-breakdown.md — Epic → subtasks decomposition
2. code-review.md — Cross-model review prompt
3. bug-fix.md — Structured debugging approach
4. documentation.md — Doc writing guidelines
5. security-audit.md — Security review checklist
6. research.md — Research task structure
7. content-creation.md — Content production workflow
8. sprint-planning.md — Sprint setup prompt
9. standup-report.md — Daily standup generation
10. lessons-learned.md — Post-task reflection

Fixed

Security

SEC-001: Path Traversal Vulnerability

• Added validation to trace and template services
• Prevents ../ path injection in file operations
• All file paths now resolved and validated against allowed directories

Performance

Telemetry Streaming:

• Large telemetry reads now streamed instead of loaded into memory
• Pagination pushed to service layer
• Optimized lookups for common queries

Quality

React Strict Mode Compliance:

• Replaced Math.random() with crypto.randomUUID() for keys
• Fixed type alignment issues
• Resolved React warning about duplicate keys

Changed

Dashboard

• Sidebar task counts now show current state, not time-filtered counts
• Archive/delete/restore operations now correctly find files on disk
• Metrics cache invalidation on status changes

Closed Issues

Issue	Title	Implementation
#82	Dev reliability (health, dev:clean, watchdog)	Health endpoints + dev scripts
#56	Dashboard filter bar with presets	DashboardFilterBar component
#53	Per-model cost tables & calculation	Cost tracking in telemetry
#51	Standup summary with model usage	/api/summary/standup
#49	Dashboard Model Usage & Cost panel	Tokens Card + Cost per Task
#48	Global usage aggregation service & API	/api/metrics/* endpoints
#47	Model Usage schema, types & API	Full telemetry system

---

[1.5.0] - 2026-02-04

✨ Highlights

• Comprehensive SOP Documentation Suite — 8 new guides covering agent workflows, sprint planning, multi-agent orchestration, code review, best practices, examples, and power user tips
• Cross-Model Code Review Enforcement — Claude ↔ GPT review gate now fully documented with RF-002 91% accuracy validation
• Bulk Archive Error Handling — Fixed silent failures in Done column archival with per-task error tracking and user feedback toasts
• Sidebar Metrics Sync — Fixed stale task counts in board sidebar by invalidating metrics cache on status changes

Added

Documentation (#US-1600)

Complete SOP Sprint with 8 new markdown files in docs/:

• GETTING-STARTED.md — 5-minute quickstart from zero → agent-ready, includes:

  • Prerequisites, manual setup wizard, UI/CLI task creation
  • Agent pickup checklist with agent-requests folder flow
  • Sanity checks section (API health, UI health, agent pickup verification)
  • Prompt registry guidance (shared resources pattern from BoardKit)
  • Documentation freshness SOP (AGENTS.md, CLAUDE.md, BEST-PRACTICES.md)
  • Multi-repo/multi-agent notes with consistent naming conventions
  • OpenClaw Browser Relay integration notes for auth-required workflows
  • Credited Neal (@nealmummau) for asking the triggering question

• SOP-agent-task-workflow.md — Complete lifecycle (claim → work → complete):

  • Roles table (PM, Worker, Human Lead)
  • Lifecycle overview with 6 stages
  • API/CLI examples for each step (start timer, status change, completion)
  • Prompt template for consistent agent instructions
  • Lessons Learned expectations + notification patterns
  • Escalation paths for blocked tasks, tooling failures, reviewer disputes

• SOP-sprint-planning.md — Epic → Sprint → Task → Subtask hierarchy:

  • Hierarchy table with real examples (MessageMeld, US-1600)
  • Sprint planner agent prompt template
  • Bulk API payload for creating entire sprints at once
  • Estimation pattern (subtasks × 0.5d = effort)
  • Assignment workflow (leave unassigned for agent pickup)
  • Example sprint (US-1600 docs sprint + RF-002 bug fix sprint)
  • After-planning checklist (recap docs, GitHub milestones, standup scheduling)

• SOP-multi-agent-orchestration.md — PM + Worker roles:

  • PM checklist (plan, assign, track, review, report)
  • Worker handoff template with clear acceptance criteria
  • Status reporting cadence (daily updates, standup summaries)
  • Error escalation paths
  • Opus-as-PM / Codex-as-worker walkthrough example

• SOP-cross-model-code-review.md — Non-negotiable opposite-model gate:

  • Enforcement rule: If Claude wrote it, GPT reviews; if GPT wrote it, Claude reviews
  • When to trigger reviews (application code required, docs optional, research optional)
  • Review workflow (create task, opposite model audits, findings as subtasks, fixes tracked)
  • Reviewer checklist (Security, Reliability, Performance, Accessibility, Docs)
  • Prompt template for consistent audits
  • Escalation paths for disagreements
  • RF-002 reference (91% accuracy validates the approach)

• BEST-PRACTICES.md — Patterns that work + anti-patterns to avoid:

  • 10 "Do This" rules (time tracking, subtasks, acceptance criteria, atomic scope, SOP updates, etc.)
  • 10 "Don't Do This" anti-patterns (no acceptance, skipping timers, grab-bag tasks, etc.)
  • Based on real-world usage with agents

• EXAMPLES-agent-workflows.md — 6 copy/pasteable recipes:

  • Feature development (BrainMeld Lessons Learned)
  • Bug fix (GH-86 bulk archive)
  • Documentation update (sanity checks in Getting Started)
  • Security audit (RF-002 style)
  • Content production (podcast clip → LinkedIn post)
  • Research & report (Champions dossiers)
  • Each includes goal, task creation, prompt, workflow steps, and deliverables

• TIPS-AND-TRICKS.md — Power user features:

  • CLI shortcuts (vk begin/done/block/unblock/time/summary)
  • Keyboard shortcuts (Cmd+K palette, arrow nav, Esc)
  • Command palette power moves
  • WebSocket awareness and polling fallback
  • MCP server setup for Claude Desktop
  • Git worktree integration patterns
  • Obsidian/Brain mirroring with brain-write.sh
  • Dev helpers (dev:clean, dev:watchdog)

• README.md — Added "Documentation Map" section listing all new guides with descriptions

Fixes

GH-86: Bulk Archive Silent Failure (#86)

Root Cause: BulkActionsBar.handleArchiveSelected() used Promise.all() with no error handling. When any single archive failed, the entire batch would silently reject with zero user feedback.

Fix:

• Import useToast hook
• Replace Promise.all() with per-task error tracking loop
• Show success toast (e.g., "Archived 5 tasks")
• Show error toast on partial/full failure with counts
• Log individual failures to console for debugging
• Clear selection regardless of outcome
• File: web/src/components/board/BulkActionsBar.tsx (+38 lines)

GH-87: Sidebar Task Counts Out of Sync (#87)

Root Cause: The sidebar uses useMetrics('24h') which polls every 30 seconds with 10-second staleTime. Meanwhile, useUpdateTask mutations did NOT invalidate the metrics cache, causing up to 30 seconds of stale data after status changes.

Fix:

• Add metrics query invalidation to useUpdateTask.onSuccess()
• Only invalidate when task status changes (prevents over-invalidation)
• Preserves timer state handling (no aggressive blanket invalidation)
• File: web/src/hooks/useTasks.ts (+9 lines)

Scripts

• scripts/dev-clean.sh — Added explicit pnpm path resolution for launchd sessions (fixes "command not found" in automated restarts)
• scripts/dev-watchdog.sh — Improved restart storm prevention with lock file + PID checking; fixed pnpm path resolution

CLI

• vk setup — New guided onboarding wizard that validates environment and helps new users get started:
  • Checks Node version (requires ≥18)
  • Verifies server is running and reachable
  • Tests API authentication
  • Optionally creates a welcome task with next steps
  • Supports --json output for automation and --skip-task to skip sample task
  • Credit: BoardKit Orchestrator (Monika Voutov) for the wizard pattern inspiration

Fixed

• Archive/Delete/Restore — Fixed "Archive failed" errors caused by filename mismatch when task titles changed after creation. Now uses findTaskFile() to locate actual file on disk by task ID prefix instead of computing filename from current title
• Sidebar Task Counts — Fixed metrics showing time-filtered counts (e.g., 33 todo) instead of current board state (e.g., 124 todo). /api/metrics/all now returns current task status counts regardless of period filter; period only applies to telemetry metrics (runs, tokens, duration)
• Backlog Count API — Fixed double-wrapped response ({data: {success, data: {count}}}) by letting responseEnvelopeMiddleware handle wrapping

Security

• SEC-001 Extended — Added path traversal validation to trace-service.ts (attemptId, taskId, traceId) and template-service.ts (templateId) using validatePathSegment() + ensureWithinBase()

Changed

• Version bumped from 1.4.1 → 1.5.0

---

1.4.1 - 2026-02-02

Security

• SEC-001 Path Traversal Prevention — added strict path segment validation + base directory enforcement in server utilities; applied to file-based services that join paths from user-controlled ids
• SEC-007 Admin Authorization — enforced admin (or admin+agent where appropriate) on mutating settings/config/activity/status-history/notifications endpoints

Fixed

• Agent Status Panel now uses real-time WebSocket updates (useRealtimeAgentStatus) and correctly handles activeAgents payloads
• Improved proxy/IP trust behavior for rate limiting (X-Forwarded-For only trusted when trust proxy is configured)

1.4.0 - 2026-02-01

✨ Highlights

• CLI Workflow Commands — Two-command task lifecycle (vk begin / vk done) plus time tracking, comments, agent status, and project management from the terminal
• Inspired by Boris Cherny's (Claude Code creator) "automate everything you do twice" philosophy

Added

CLI Workflow Commands (#44)

• Composite workflows — Complete task lifecycle in single commands:
  • vk begin <id> — Sets in-progress + starts timer + updates agent status to working
  • vk done <id> "summary" — Stops timer + sets done + adds comment + sets agent idle
  • vk block <id> "reason" — Sets blocked + adds comment with reason
  • vk unblock <id> — Sets in-progress + restarts timer
• Time tracking CLI — Full time management from terminal:
  • vk time start <id> — Start time tracker
  • vk time stop <id> — Stop time tracker
  • vk time entry <id> <seconds> "description" — Add manual time entry
  • vk time show <id> — Display time tracking summary (total, running status, entries)
• Comments CLI — vk comment <id> "text" with optional --author flag
• Agent status CLI — Manage agent presence:
  • vk agent status — Show current agent status
  • vk agent working <id> — Set to working (auto-fetches task title)
  • vk agent idle — Set to idle
  • vk agent sub-agent <count> — Set sub-agent mode with count
• Project management CLI — vk project list and vk project create "name" with --color and --description flags
• All commands support --json for scripting and automation
• 5 new command modules, 18 subcommands, 651 lines added

---

---

1.3.0 - 2026-02-01

✨ Highlights

• GitHub Issues Bidirectional Sync — Import issues with the kanban label and push status changes back to GitHub
• Activity Feed — Full-page chronological activity feed with filtering, real-time updates, and compact/detailed toggle
• Daily Standup Summary — Generate standup reports via API or CLI with completed, in-progress, blocked, and upcoming sections

Added

GitHub Issues Sync (#21)

• GitHubSyncService (464 lines) with polling, label-based field mapping, and circuit breaker
• Inbound: import issues with kanban label as tasks
• Outbound: push status changes (done → close issue, reopen on todo/in-progress/blocked) and comments
• Label mapping: priority:high → priority field, type:story → type field
• Config: .veritas-kanban/integrations.json, state: .veritas-kanban/github-sync.json
• TaskGitHub interface in shared types: {issueNumber, repo, syncedAt?}
• New API endpoints:
  • POST /api/github/sync — trigger manual sync
  • GET /api/github/sync/status — last sync info
  • GET /api/github/sync/config — get config
  • PUT /api/github/sync/config — update config
  • GET /api/github/sync/mappings — list issue↔task mappings
• New CLI commands: vk github sync, vk github status, vk github config, vk github mappings

Activity Feed (#33)

• Full-page chronological activity feed accessible from header nav (ListOrdered icon)
• agent field added to Activity interface
• ActivityFilters for combinable filtering (agent, type, taskId, since, until)
• GET /api/activity enhanced with query params: ?agent=X&type=Y&taskId=Z&since=ISO&until=ISO
• GET /api/activity/filters — distinct agents and types for filter dropdowns
• ActivityFeed.tsx component with day grouping, 15 activity type icons, filter bar, compact/detailed toggle
• Infinite scroll via IntersectionObserver
• Real-time WebSocket updates
• ViewContext for board ↔ activity navigation

Daily Standup Summary (#34)

• GET /api/summary/standup?date=YYYY-MM-DD&format=json|markdown|text
• Sections: completed, in-progress, blocked, upcoming, stats
• generateStandupMarkdown() and generateStandupText() in SummaryService
• CLI: vk summary standup with --yesterday, --date YYYY-MM-DD, --json, --text flags
• 12 new tests

Changed

• MAX_ACTIVITIES increased from 1,000 to 5,000

---

1.2.0 - 2026-02-01

✨ Highlights

• Standardized API Response Envelope — All endpoints return a consistent {success, data, meta} format with typed error classes
• Abstract File Storage — Repository pattern decouples services from the filesystem
• Blocked Task Status — Full support for blocked tasks across MCP, CLI, and board

Added

Standardize API Response Envelope (#2)

• 4 new error classes: UnauthorizedError, ForbiddenError, BadRequestError, InternalError (in middleware/error-handler.ts)
• sendPaginated(res, items, {page, limit, total}) helper for pagination metadata in envelope
• Response envelope format:
  • Success: {success: true, data, meta: {timestamp, requestId}}
  • Error: {success: false, error: {code, message, details?}, meta}
  • Pagination: meta includes {page, limit, total, totalPages} on paginated endpoints

Abstract File Storage (#6)

• 5 new repository interfaces: ActivityRepository, TemplateRepository, StatusHistoryRepository, ManagedListRepository, TelemetryRepository
• StorageProvider extended with new repositories
• fs-helpers.ts — centralized filesystem access (only file that imports fs)

Blocked Task Status (#32)

• MCP tools Zod/JSON schema definitions updated for blocked status
• MCP active tasks filter updated to include blocked
• CLI help text updated
• CLI status color: blocked = red

Changed

• All 11 route files standardized — zero ad-hoc {error: "..."} patterns
• Auth middleware errors standardized to use typed error classes
• All 10 services migrated off direct fs imports to use fs-helpers.ts

---

1.1.0 - 2026-01-31

✨ Highlights

• Built-in Chat Interface — Talk to AI agents directly from the board or any task, with streaming responses and markdown rendering
• Agent Routing Engine — Tasks auto-route to the best available agent based on type, project, and capabilities
• Agent Selection on Task Creation — Choose which agent handles a task when you create it
• Hardened Infrastructure — Rate limiting, circuit breakers, file locking, request timeouts, data integrity checks, and more

Added

Chat Interface (#18)

• Full chat panel accessible from any task or the board header
• Streaming AI responses with real-time WebSocket delivery
• Floating chat bubble with pulse indicator for new messages
• Chat sessions stored as markdown files with YAML frontmatter
• Gateway integration for AI responses via Clawdbot
• Chat export as markdown (download icon in header)
• Clear chat history with confirmation dialog
• Mode toggle: Ask (read-only queries) vs Build (changes, files, commands)
• Keyboard shortcut support
• Auto-focus input after sending messages
• Tool call display with expandable input/output sections

Agent Routing Engine (#16)

• Task-aware routing that matches tasks to agents by type, project, and capabilities
• Routing rules configurable per agent in Settings → Agents
• API endpoints for routing queries and rule management
• Full test coverage (17 tests)

Agent Selection on Task Creation (#17)

• Agent dropdown in the Create Task dialog
• Auto-routes to best agent based on task type, or allows manual override
• Agent field displayed in task metadata section

Agent CRUD Management

• Full Add/Edit/Remove for agents in Settings → Agents
• Add Agent form with name, type slug (auto-generated), command, and args
• Edit/Remove via inline icons (default agent protected from deletion)
• AgentType loosened from fixed enum to any string slug — fully custom agents

Board Filter: Agent

• Filter board by assigned agent in the FilterBar
• Agent indicator dots on task cards match filter state

Infrastructure & Security

• Rate Limiting — Per-route tiered thresholds (auth, API reads, writes, uploads)
• Circuit Breaker — Automatic failure detection for external service calls with configurable thresholds
• File Locking — FIFO queue prevents race conditions on concurrent file writes
• Request Timeouts — Middleware kills hung connections before they pile up
• Data Integrity — Hash-chain verification + automatic backup on startup with rotation
• Audit Log — Immutable hash-chain audit trail for sensitive operations
• Health Endpoint — Liveness, readiness, and deep checks (storage, disk, task file)
• API Envelope — Standardized { success, data, meta } response format across all endpoints
• Schema Validation — Zod schemas on all mutating API routes
• Metrics — Prometheus-compatible /metrics endpoint for monitoring
• WebSocket Heartbeat — Connection keep-alive with automatic reconnection and connection limits
• Error Boundaries — React error boundaries with graceful fallback UI
• Dependency Audit — Automated vulnerability scanning in CI

Storage & Architecture

• Abstract file storage behind repository interface (prep for future database backends)
• Structured logging with pino (replaced all console.* calls)

First-Run Experience

• Example tasks auto-populate the board on first run (4 sample tasks)
• Manual seed script: pnpm seed
• Task data .gitignored — your data stays private

Dark/Light Mode

• Settings → General → Appearance toggle (moon/sun icon)
• Persists to localStorage; default is dark mode
• Inline script prevents flash of wrong theme on load

UI Theme

• Primary color: purple (270° 50% 40%) with white text
• Focus rings, switches, and accents updated to match

Documentation

• TROUBLESHOOTING.md with common issues and solutions
• Comprehensive FEATURES.md reference
• Agentic AI Safety best practices guide
• Roadmap section linking to v1.1 milestone
• Competitive comparison table
• OpenClaw (formerly Moltbot/Clawdbot) attribution updated

Per-Status Selection (#24)

• Select All checkbox per column header
• Toolbar buttons for bulk operations scoped to selected status
• Column checkboxes for quick multi-select

Fixed

• Chat delete not clearing UI — React Query kept stale cached data after session file was deleted; now uses removeQueries to nuke cache
• Chat send broken after delete — Server now recreates task-scoped sessions instead of throwing 404
• Cross-column drag-and-drop — Tasks reliably move between columns with local state management during drag
• Dashboard agent comparison — Fixed broken data fetch (raw fetch → apiFetch for envelope unwrapping)
• Dashboard drill-down — Removed duplicate X button, fixed focus ring clipping, wired up open-task event
• Localhost auth rate limit (#25) — Exempted localhost from rate limiting
• Numeric inputs — Clean inputs without browser spinners (#19)
• Timer start/stop — Optimistic UI toggle + cache patch for instant feedback
• Task cache fragmentation — All routes now use TaskService singleton
• Sprint/Agent label alignment — Fixed form layout in task detail panel
• Sticky header — Fixed positioning + matched indicator dot sizes
• Keyboard test infinite loop — Resolved render loop in useKeyboard + memoized context
• Agent idle timeout — Increased from 5 to 15 minutes to reduce false resets
• File lock ordering — Added in-process FIFO queue for deterministic write ordering
• Search filters — Added task ID to board and archive search

Changed

• Agent status popover: moved idle description to bottom, added activity history link
• WebSocket indicator: click popover with connection status explanation
• Dashboard layout: Daily Activity (75%) + Recent Status Changes (25%) side-by-side
• Rolling average line: cyan-teal to contrast purple theme
• Bar chart hover: subtle muted fill instead of white flash
• All repo links updated to BradGroux (primary repo)
• All contact emails standardized to contact@digitalmeld.io
• Test suite: 72 files, 1,270 tests (up from 61 files / 1,143 tests)

---

1.0.0 - 2026-01-29

🎉 Initial Public Release

Veritas Kanban is an AI-native project management board built for developers and autonomous coding agents.

Features

Core Board

• Kanban board with drag-and-drop between columns (Backlog, To Do, In Progress, Review, Done)
• Task detail panel with full editing (title, description, priority, status, type, project, sprint)
• Subtasks with progress tracking on cards
• Task type system with icons and color-coded borders
• Sprint management with auto-archive
• Bulk operations and keyboard shortcuts

Code Workflow

• Git worktree integration for code tasks
• Diff viewer for code review
• Line-level review comments
• Approval workflow with review decisions
• Merge and close integration

AI Agent Integration

• Agent orchestration system for autonomous task execution
• Agent status tracking (idle, working, sub-agent mode)
• Time tracking per task with automatic and manual entries
• REST API designed for AI agent consumption
• MCP (Model Context Protocol) server for LLM tool integration
• CLI for headless task management

Dashboard & Analytics

• Sprint velocity tracking
• Cost budget tracking with daily digest
• Task-level metrics and telemetry
• Status history timeline

Security

• JWT authentication with secret rotation
• Admin key + API key authentication
• CSP headers with Helmet
• Rate limiting with express-rate-limit
• CORS origin validation
• WebSocket origin validation
• Server-side MIME type validation for uploads
• Markdown sanitization (XSS prevention)
• Timing-safe credential comparison
• Credential redaction from task data

Performance

• In-memory task caching with file watchers
• Config caching with write invalidation
• Gzip response compression
• Lazy-loaded dashboard with vendor chunk splitting (69% bundle reduction)
• Pagination and summary mode for large datasets
• Reduced polling when WebSocket connected
• Telemetry retention and automatic cleanup

Infrastructure

• Production Dockerfile with multi-stage build (runs as non-root)
• GitHub Actions CI pipeline
• Pre-commit hooks with husky + lint-staged
• Structured logging with pino
• Request ID middleware for tracing
• Graceful shutdown with service disposal
• Unhandled rejection and exception handlers

Documentation

• OpenAPI/Swagger API documentation
• Deployment guide (Docker, bare metal, nginx, Caddy, systemd)
• Security audit reports
• Contributing guide with conventional commits
• Code of Conduct (Contributor Covenant v2.1)

Testing

• 61 test files, 1,143 unit tests (server + frontend) with Vitest
• End-to-end tests with Playwright (19/19 passing)
• Gitleaks pre-commit hook for secret scanning

Technical Details

• Frontend: React 19, Vite 6, TypeScript 5.7, Tailwind CSS 3.4, Shadcn UI
• Backend: Express 4.21, TypeScript, file-based storage
• Testing: Playwright 1.58, Vitest 4
• Runtime: Node.js ≥ 22.22.1, pnpm ≥ 11.0.0

---

Built by Digital Meld — AI-driven enterprise automation.