Stage billing previously counted only the parent session's own LLM
calls. SubAgentResult carried no usage, so when a stage completed, the
authoritative billing replaced the live event-accumulated total with a
main-session-only number and every child session's spend vanished. On a
real run this collapsed a ~$175 total to ~$10.
Fix, part 1 — child spend reaches the parent's totals:
- SubAgentSupervisor keeps a spend ledger (SessionSpend). The child
runner records last_input_usage/last_input_cost after every
generation, including failed turns.
- The parent session drains the ledger into its input totals before
publishing them, so AgentApiBackend::record_input_usage bills the
whole session tree. Nested subagents compose: grandchildren drain
into the child, whose runner records into the parent.
Fix, part 2 — the live number rides the same data:
- Each session emits a monotonic lifetime counter of its own spend
(session_spend on AssistantMessage, session_billing on agent.message
events, spec'd in fabro-api.yaml).
- The projection keeps the latest counter per session at the run level
and attributes only the delta to the stage a message belongs to, so
dropped or replayed events can no longer lose or double-count spend,
and sessions reused across stages (shared threads) bill each stage
for exactly its own turns. Legacy events without the counter fall
back to per-message summing.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Structural cleanup of the durable pull request creation feature, from a
three-agent review (reuse, quality, efficiency) of the branch:
- Move the supervisor out of handler/ into server/pull_request_supervisor.rs,
collapse its double bookkeeping into one task-id map, and fold the five
copy-pasted failure arms into attempt_pull_request_creation.
- Tag pull_request.failed events with the creation id they resolve, so a
publish-stage failure can never fail an unrelated explicit creation. The
reducer gains PullRequestCreation::succeed/fail transition methods.
- Scan pending creations through a narrow projection-cache accessor instead
of materializing every run summary, raise the scan interval to 30s (notify
covers the live path), and cap retries for runs whose worker cannot even
record a failure.
- Answer "creation already pending" POSTs before taking the per-run create
lock, which a worker can hold for the whole creation.
- Replace the hand-rolled per-run lock map with fabro_store::KeyedMutex.
- Reuse cheap Arc'd projections (cached_run_projection) on the poll endpoint
and in the worker instead of deep-cloning run summaries and diffs.
- Merge ExistingPullRequest into fabro_github::CreatedPullRequest and
extract one reconcile_existing_pull_request helper for both call sites.
- Give the client poll loop a 15-minute deadline; document that Retry-After
and the poll interval are the same constant.
- Resolve a wedged pending creation (run already has a pull request) as a
durable failure instead of skipping it forever.
- Tests: shared wait_for_pull_request_creation helper, a pinned generation-
failure assertion, and a new pipeline test proving reconciliation adopts
an existing PR without an LLM call or create request.
Verified: cargo build --workspace, cargo nextest run --workspace (7,767
passed), nightly clippy -D warnings, fmt --check, insta (no pending), bun
typecheck in fabro-api-client.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
250ms was too aggressive: a server that was reachable but slightly slow
to answer /health made `fabro doctor` report a failed health check.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The manifest builder's best-effort pre-run push converted every result
into a PreRunPushOutcome that was serialized into GitContext, expanded
into five OpenAPI union arms, and generated into API clients — but no
production path ever read it; every field read was a test.
Delete the concept while preserving the behavior:
- Drop the PreRunPushOutcome enum and GitContext.push_outcome from
fabro-types; GitContext keeps origin_url, branch, optional sha, and
dirty, which remain real execution inputs and provenance.
- Rename the manifest outcome builder to push_manifest_branch_best_effort,
a side-effect-only helper with the same decision rules: skip without an
origin, skip on configured-repository mismatch, skip when the branch is
already synced, otherwise push noninteractively and discard the result
without failing manifest creation or logging raw Git stderr.
- Prove the push through repository state instead of the deleted enum: a
branch ahead of a local bare origin is pushed during manifest build, a
mismatched configured repository is not, and a failing remote helper
still cannot fail manifest creation.
- Remove push_outcome from GitContext in OpenAPI, delete the five-arm
union schemas, and drop the fabro-api type replacement and re-export.
- Keep one regression proving historical run.created events with a nested
push_outcome still deserialize through ordinary unknown-field tolerance
and reserialize to the reduced shape. No migration or event rewrite.
Old JSON carrying the removed field stays readable. Newly generated
clients omit a field older servers required, so new-client-to-old-server
compatibility is intentionally not promised for this pre-1.0 contract.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Drop ManifestTarget.identifier (the raw token the user typed) and
ManifestGoal.path (the original goal-file path) from the OpenAPI
manifest schema, the Rust manifest builder, the regenerated Rust and
TypeScript client types, and every canonical test fixture. Neither
field had a production reader: the server selects the workflow by
target.path and consumes only the resolved goal type and text.
Target path, goal type/text, manifest versioning, and submitted-byte
persistence are unchanged. Old request bodies that still carry the
removed properties remain accepted through unknown-field tolerance,
pinned by a dedicated public-route regression test.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
GitHub repository slug and git ref selector syntax now has one owner:
fabro-types::repository defines GitHubRepositorySlug with a try_new
constructor and the is_valid_github_ref_selector predicate.
fabro-automation keeps its public type path as a re-export of the same
type and delegates its existing parser and ref validation to the shared
grammar, preserving its exact error variants and messages. Server
checkout and materialization code imports the type from its canonical
owner. No wire, API, or behavior change.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
AgentPermissions duplicated fabro_types::PermissionLevel: same variants,
same kebab-case wire form, same crate. PermissionLevel is strictly richer
(Hash, strum, clap::ValueEnum) and is already the with_replacement target
for the OpenAPI PermissionLevel schema, whose values are identical to the
AgentPermissions schema this branch deletes.
Delete AgentPermissions and type the [cli.exec.agent] permissions setting
as PermissionLevel. This drops the adapter match in `fabro exec` and the
`as AgentPermissionLevel` alias that existed only to tell the two names
apart. The TOML wire form is unchanged.
Removing run.agent.permissions also changed the serialized run spec, but
two fabro-cli inline snapshots still carried "permissions": null. They
failed on this branch and passed on main. Accept the updated snapshots.
Also tighten the removed-setting test to assert the exact unknown-field
message, rename its module to run_agent now that it covers more than
fabro_tools, and drop three doc references to the removed setting.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Review pass over the reuse change. No intended behavior changes.
- share one definition of the initial generation from fabro-types instead
of three copies across fabro-types, fabro-agent, and the supervisor
- give each child one SubAgentHandle instead of threading the supervisor's
state, callback, and notification sender through five functions, and
collapse the repeated signal-then-drain pairs into publish()
- move `reusable` inside SubAgentStatus::Finished so a closed agent can no
longer be marked reusable
- clear the lifecycle draining flag with an RAII guard, so one panicking
callback cannot silence every later lifecycle event
- tear down a session that failed to initialize right away rather than
holding it and its sandbox until the parent closes the agent
- look agents up through SupervisorState::agent/agent_mut instead of five
copies of the same not-found error
- drop the unreachable cleanup_started branch and the test-only emit_event
whose only caller was its own test
- render subagent starts from one ProgressEvent and one display method,
deriving the spawn/turn distinction from the generation
- set projected subagent status through one helper instead of four
identical reducer arms
- drive the generation-pinned wait test through spawn/send_input rather
than hand-writing private supervisor state
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Resolve the model catalog table conflict in docs/public/core-concepts/models.mdx
by keeping both changes: this branch's `kimi` -> `moonshot` provider rename for
the Kimi rows, and main's new DeepSeek V4 rows.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The preamble removal changed the `stage.completed` payload, so the
`attach --json` inline snapshot no longer matched. Drop the stale
`current.preamble` line.
Reword the `context_values` doc row. `stage_context_values` only strips
runtime-only keys; it does not normalize artifact pointers to blob refs
the way `artifact::durable_context_snapshot` does, so calling it a
durable snapshot overstated it. Point readers at `checkpoint.completed`
for the durable projection.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Two artifacts can share a filename, a retry, and an absent stage, in
which case the winner was whichever the object store listed first. Break
the tie on the serialized stage ID, which is the third key the artifacts
page sorts on. Compare the `node@visit` string rather than StageId's own
ordering: the page compares the string, so "unknown@2" beats
"unknown@10" there and now here too.
The spec said captures from the `start` and `exit` nodes are excluded,
but the exclusion is by handler type, so a node named `start` that does
real work keeps its artifacts. Say that instead.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Raise the replacement test's deadline to 20s. The server takes up to 1s to
notice the replacement and then bounds its own shutdown at 5s, so the old
5s deadline sat below the worst case and could fail a healthy server on a
loaded runner. A passing run still exits in about a second.
Reword the SHUTDOWN_TIMEOUT comment.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Deflate flushes its output in 8 KiB blocks, and each write became its own
allocation, channel send, and HTTP body frame. A 64 KiB BufWriter in
front of the sink cuts all three by eight.
An artifact deleted between the listing and its read no longer aborts the
whole archive. That race is a run being pruned mid-download; leaving the
file out beats handing back a truncated ZIP missing everything after it.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Exit the process from `main` for every command instead of returning. The
`mcp start` command parks Tokio's stdin reader on a read only the MCP host
can end, so dropping the runtime waits forever. Exiting in `main` also keeps
the CLI telemetry event, which the previous exit inside the MCP command
skipped.
That removes the reason for the `McpServerExit` enum, whose only job was to
carry an implementation detail out to the CLI so it could exit.
Watch the executable through its device and inode on Unix. That is a
complete file identity, so the length and modification time no longer add
anything. Drop the PATH scan: `current_exe` reports the symlink itself on
macOS, so it detects a Homebrew relink without it. This also drops the
`fabro-static` dependency and a clippy suppression.
Bound the shutdown wait after an upgrade is detected. The transport closes
by writing to a stdout the host may already have stopped reading, which
could hang the exit the change is supposed to trigger.
Log a warning when upgrade detection cannot start, rather than disabling it
silently.
Share one spawn helper between the two raw stdio tests, and link the test
executable instead of copying 200 MB of binary.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Path safety now lives in one place. The NUL-byte and drive-letter rules
move from a server-only helper into the store's own filename validation,
so uploads reject those paths at write time instead of only the ZIP read
path catching them. The download still re-checks, because artifacts
stored before the rule existed can still carry an unsafe path, but it
now skips a bad path rather than failing the whole archive.
Promote is_boundary_stage to RunProjection and drop the three identical
private copies. The ZIP download used a node-name match instead, which
would have dropped artifacts from a working node that happened to be
named "start".
Compress the archive. Entries were Stored while the response was also
excluded from transfer compression, so text artifacts moved at full
size. async_zip gains the deflate feature; async-compression and flate2
were already in the lock file.
Log archive failures unconditionally. The send-succeeded guard meant a
client that had already disconnected left no record at all, which is the
case where the log is the only evidence.
Also: collapse the duplicate 500 arms, drop the dead stage-ID tiebreaker
and the cached order in the selection map, name the accessible label
after the visible one, share the run URL prefix between the two download
href builders, and document the mid-stream truncation behavior in the
OpenAPI description.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Consolidation pass over the fallback feature, no intended behavior
changes beyond noted validation and event-shape cleanups:
- Unify the two parallel notice types: FallbackPlanNotice is gone;
ModelFallbackNotice now owns the runtime NoNearbyReasoningLevel case
and the shared ChainEmpty wording. Notices emit through a new
Emitter::notice_scoped with their own level, and each distinct notice
is emitted once per run instead of on every LLM call.
- Move canonical_model_id onto Catalog so chain keys are written and
read through one function; reject provider-qualified fallback keys,
which could never match at dispatch and were silently dead config.
- Type FallbackTarget as ProviderId/ModelId, removing repeated
ProviderId::new re-wrapping at every use site.
- Derive FallbackPlan's current route from a position index instead of
storing current/requested_controls copies; advance() no longer has
unreachable None branches.
- Bundle the agent invocation's live state (session, bridge, lease,
forwarder, accounting) into LiveAgentInvocation; failover_agent_session
drops from 21 parameters to 7 and the six copies of the
abort/discard/classify teardown collapse into two methods.
- Share one route_request builder between one_shot and its failover
loop; complete_one_shot_request takes the request by value instead of
deep-cloning the message payload per call.
- Event::Failover carries FailoverProps directly; the props' original
route and attempt fields are now required, and reasoning efforts are
typed ReasoningEffort instead of strings.
- Reuse RunModelSettings/RunModelControls in fabro-api via
with_replacement, add the missing controls property to the OpenAPI
schema, regenerate the TS client, and add the type-identity/JSON
parity test.
- Smaller cleanups: ReasoningEffort::closest_supported uses enum
discriminants; ModelFallbackPolicy gains len(); resolve_model_fallbacks
takes a provider slice; duplicate-target filtering lives only in the
resolver.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
