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>
Addresses review feedback on the fallback notice work.
A provider-only fallback such as `openrouter` needs the primary model's
catalog entry to find the closest capability match. When the primary is
itself a passthrough selector there is no entry, so every provider-only
candidate was skipped with "provider `X` has no compatible model" even
when that provider had plenty. Adds a `PrimaryNotInCatalog` notice that
names the missing primary instead of blaming the provider.
Also from review:
- `code()` was a wildcard fallthrough, which docs/internal/events-strategy.md
forbids for new variants. Now exhaustive.
- `NoConfiguredOffering` discarded the `providers` list that
`NoEligibleOffering` hands it. The notice now names the providers that do
offer the model.
- `ModelFallbackNotice::reference` was a rendered `String`; it is now the
`ModelRef` it came from, which also drops the per-candidate double
allocation the previous refactor introduced.
- `ResolvedStartLlm` unpacked and repacked `ResolvedFallbackChain`
field-for-field; it now holds it directly.
- Added `FallbackTarget: Display` as `provider:model`, replacing two
hand-written `"{}:{}"` format strings.
- `Catalog::select` still inlined the `require_provider` body.
- Emission moved to `ModelFallbackNotice::emit_all`, covered by a new test
proving notices reach the event stream with the right level, code, and
message. Nothing tested that hand-off before.
Documented in `resolve_fallback_chain` why an unknown provider stays a hard
error while an unconfigured one is skipped, and that an unqualified unknown
selector pins to the primary's provider.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Follow-up cleanup on the portable fallback chain work.
- Add `Catalog::require_provider` and `Catalog::provider_id`, replacing the
`catalog_provider_id` free function in `start.rs` and two copies of the same
`provider(..).ok_or_else(UnknownProvider)` block inside the catalog.
- Add `FallbackTarget::new` and use it for the six struct literals that each
stringified a provider and model by hand.
- Extract per-candidate resolution into `resolve_fallback_candidate`, returning
a `FallbackCandidate` that is either a target or the skip reason. This flattens
`resolve_fallback_chain` from four levels of nesting to one loop and splits the
qualified/unqualified model arms into separate match patterns.
- Drop the `seen` HashSet and its per-candidate key clones in favor of a
`contains` check on the chain being built; fallback chains hold a handful of
entries.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
One row per separator rule, so a regression names the input that broke
instead of pointing at a combined fixture string. Also record why
`add_class` keeps insertion order: `fidelity` falls back to the first class
for the thread ID.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Replace the hand-rolled byte scanner in strip_css_comments with a
str::find loop over "/*" and "*/".
Drop the quote and backslash tracking. The stylesheet language has no
string literals: parse_declarations ends a value at the first ';' or
'}' with no quote awareness, and values flow into AttrValue::String
verbatim, so a quoted model name is just an unknown model. Tracking
quotes here also created a failure mode the simple scan does not have.
An unpaired apostrophe, as in `model: don't`, disabled comment
stripping for the rest of the input and then blamed a well-formed
comment for the parse error.
Also drop the Cow and its copied_through watermark. They avoided one
allocation on a graph attribute of a few hundred bytes, parsed once per
workflow load, in a function whose caller already clones the attribute
and whose parser allocates a String per property and per value.
Extract excerpt() for the error snippets. The two existing call sites
sliced raw bytes at index 20, which panics when a multi-byte character
straddles the cutoff; model_stylesheet is arbitrary user text, so that
was reachable.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Node classes were built in two places. The parser split the `class`
attribute on commas and whitespace, but the import transform re-split the
raw attribute on commas only. A space-separated class on an import
placeholder became a single class name, so stylesheet rules did not match.
That included the `class="fast shared"` example in the imports docs.
- add `Node::add_class`, replacing the duplicate append helpers in
`SemanticState` and `ImportTransform`
- read `node.classes` in `placeholder_config` instead of re-parsing the raw
attribute, so class splitting happens in exactly one place
- name the separator rule `split_class_attr`, splitting on commas and then
whitespace so empty entries need no trimming
- drop the unused `Node::class` accessor that invited the re-parse
- keep the comma-compatibility note in the DOT attribute reference only
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The 10 MiB cap on resolved stdin_source values is tight for wide
fan-in: a context.parallel.results batch from a large for_each round
carries tens of structured agent outputs, and a merge step that feeds
them to a deterministic command hits the ceiling as a hard
deterministic failure. Raise the ceiling to 30 MiB; it still bounds
peak memory and remote uploads, just with headroom matched to the
fan-out sizes for_each already allows.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Workflow agent sessions run at PermissionLevel::Full with the whole
tool registry exposed, so blocking pre_tool_use hooks are the only
policy boundary they have. The child-session factory built for
spawn_agent dropped tool_hooks from the child's SessionOptions, so a
subagent's tool calls never reached the run's hooks: any agent that
could spawn a subagent got an unguarded read-write-shell escape from
every hook-enforced policy.
Clone the parent's tool_hooks into the factory, the same way the
permission level is already carried, so child sessions inherit the
parent's hook boundary. The new end-to-end test drives the real
create_session path against a scripted mock provider: the parent
spawns a child, the child executes read_file, and the hooks must see
both the parent's spawn_agent and the child's read_file.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
CommandHandler also serves type="tool" nodes, so the runtime message now
says "Node '...'" to match the stdin_source_valid lint wording.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- ExecStreamingRequest: drop #[non_exhaustive] and the six Option-taking
builder setters; call sites use struct literals over ::new(), matching
GrepOptions/WalkOptions, and providers can destructure exhaustively
- Docker: pass ExecStreamingRequest through docker_exec_shell_streaming
instead of seven positional args; revert the no-op StartExecOptions
- Daytona: stdin temp-file cleanup is now best-effort (mirrors
DaytonaSession::close) so a failed delete cannot fail a completed
command or double-delete from Drop; upload overlaps session creation;
one shared DAYTONA_CLEANUP_TIMEOUT
- write_process_stdin tolerates ConnectionReset/ConnectionAborted so a
command that stops reading stdin does not fail on TCP Docker daemons
- Local sandbox aborts the stdin writer after process exit instead of
joining unbounded
- Cap stdin_source payloads at 10 MiB, mirroring the for_each bound
- Add Node::context_key_attr() tri-state so the handler and lint rule
share one definition of a valid context-key attribute
- inert_attribute canonicalizes handler types via StageHandler, fixing
false warnings for command attrs on tool nodes
- Share resolve_flat_context_value between command stdin and for_each;
resolve_json_value takes Value by value, removing a deep clone
- Reuse MockSandbox in command handler stdin tests instead of extending
SpySandbox with a hand-rolled streaming override
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Addresses a Copilot review comment on #653.
The source array is runtime data, usually produced by a model, so its
length is not something a workflow author reviewed. Two changes, so an
over-long array degrades into a clear error rather than memory pressure.
Cap the item count at 1000. Above that the stage fails deterministically
before `parallel.started`, alongside the other for_each contract
violations, and the message says how to reduce the array.
Fork the parent context inside the branch task, after it acquires a
`max_parallel` slot, instead of at dispatch time. Live context copies now
track `max_parallel` rather than item count. Only the branch's own
preamble entry is moved into the task, so the shared stash is not cloned
per branch either.
The reviewer also suggested replacing spawn-all with `max_parallel`
workers pulling from a queue. Not done here: with the fork deferred, a
pending task holds little beyond its item, and reshaping the dispatch
loop would change cancellation and scope-reservation ordering, which
deserves its own review.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Addresses three Copilot review comments on #653.
`item_label` comes from a model or a workflow author, and it reaches the
terminal through the CLI progress display. A label could carry ANSI
escapes, newlines, or bidi overrides and rewrite what the operator sees.
It could also be whitespace-only, giving a branch a blank identity.
Add `text::sanitize_display_label`: strip ANSI sequences, drop control
and bidi-reordering characters, trim, and elide past 80 characters.
Return an empty string when nothing printable survives so callers fall
back to an identity they control.
Apply it where the label is created, so events, the store, and the web
UI all get a clean value instead of each consumer having to remember.
`parallel_branch_display` sanitizes again, because a run recorded before
this commit still has raw labels in its event log.
`emit_branch_retrying` now sets `stage.retrying`'s `index` from the
branch stage's execution ordinal, matching the envelope `stage_id` and
the meaning every other emitter gives that field. The branch's position
in the fan-out is already on `parallel.branch.started`.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
build_branch_plan read the runtime array before run_branches looked at
`simulated`, so every for_each workflow failed under --dry-run with
"for_each source '...' was not found in workflow context". Nothing had
populated the key yet: upstream LLM nodes take Handler::simulate, which
returns no context updates.
A dry run now stands in one placeholder item when the source is absent or
unusable, and simulates the template target once. Graph-shape mistakes
still fail, since catching those is the point of a dry run.
Also from review:
- ITEM_FENCE_PREFIX replaces the bare "untrusted-" literal that
render_item_data and its test each spelled out.
- ItemRecordingHandler no longer guesses an item label by substring
search. Nothing asserted it, and the third item's label "2" matched
stray hex from the random fence tag about two thirds of the time.
- The twin test reads keys::PARALLEL_RESULTS instead of the raw string.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Simplification pass over the for_each branch. No behavior change.
Share what was duplicated:
- Node::prompt_or_label replaces the "prompt, else label" fallback that
agent, prompt, and the for_each item injector each wrote out.
- context::lookup_flat replaces the "exact key, then strip context."
lookup that condition.rs had twice and the for_each source had again.
- is_llm_handler_type replaces the inline agent/prompt match, so the
runtime and the for_each_contract rule agree by construction.
- find_join_node now takes branch ids, so a for_each fan-out passes its
template target instead of needing find_join_for_target.
- collect_events moves to test_support; parallel and integration tests
shared one copy already.
- One ScriptedHandler replaces four test handlers that differed only in
what they returned.
Straighten the branch retry loop:
- Reserve the branch scope once before the loop instead of guarding it
with an Option, which removes three expect() calls.
- acquire_branch_permit and backoff_or_cancel replace the cancel-aware
select! blocks the loop repeated verbatim.
- Keep the match arms in Executor::execute_with_retry order so the two
loops stay easy to compare.
Drop redundant state:
- BranchPlan::is_for_each derives from template_target_id.
- for_each_contract checks node type before source shape, so one
mistake reports one diagnostic.
Prove the new wire fields survive the OpenAPI boundary: the fabro-api
round-trip fixture now carries index and item_label.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The schema advertises defaults for `block` and `timeout`, but the
executor errored when either was absent. Apply the advertised defaults
instead, and keep the type check for values that are present.
The required list stays as the Claude 5 contract declares it. Constants
now hold the defaults and the maximum so the schema and the executor
cannot drift.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Reduce duplication and over-specification introduced with the Modal
provider, without changing shipped behavior.
- Extract enabled_provider_catalog and assert_deep_tool_round_trip in
the fabro-llm integration tests. The Poolside, Fireworks, OpenRouter,
and Modal deep round trips were four near-identical copies.
- Add ApiCredential::with_extra_headers for providers that authenticate
with request headers instead of an API key.
- Replace the unreachable require_env guards in the Modal e2e test with
the std::env::var form used by every sibling test, and register
MODAL_TOKEN_ID and MODAL_TOKEN_SECRET in EnvVars.
- Collapse modal_requires_both_vault_proxy_tokens to a single case. The
loop rebuilt the whole built-in catalog per iteration.
- Drop tautological and over-specified assertions: the api_key_url doc
URL, the forced default/probe lookups on a single-model provider, and
the get_on_provider loop that could not fail.
- Inline the single-use modal_env_catalog fixture and note why it
overrides the shipped secrets templates.
- Sort the MODAL_* keys in .env.example, and record in modal.toml why
api_id keeps the Hugging Face capitalization.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Conflicts were between this branch's parallel-branch identity work and
main's stage billing, review targets, and live stage timing.
- Stage fixtures: main added `billing` to each per-file `makeStage`; this
branch had hoisted one builder into `lib/test-utils`. Kept the hoisted
builder and gave it `billing: makeBilledTokenCounts()`, so both intents
hold and the field list stays in one place. `stage-sidebar.test.ts` also
builds raw `RunStage` wire payloads, so it keeps importing
`makeBilledTokenCounts` directly.
- Import lists (`run_projection.rs`, `fabro-api/src/lib.rs`,
`run_state.rs`, `stage_projection_round_trip.rs`): unioned both sides —
`ParallelBranchId` alongside `timing`, `ReviewTarget`,
`ReviewTargetKind`, `AttrValue`, `Node`, and
`StageToolBatchProjection`.
- `fabro-server` tests: git interleaved two unrelated new tests into one
body. Split them back into
`list_run_stages_exposes_parallel_branch_identity` and
`run_billing_includes_live_stage_timing_in_rows_and_totals`.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Restore the documented SDK env credential facade without reintroducing run fallback behavior. Fail closed on GitHub permission resolution, require worker storage at the CLI boundary, and align interpolation names and generated docs.
Branches bypass the engine's stage.started/stage.completed lifecycle, so
no SWR key invalidated the stages list while a fork ran. The new live
branch rows stayed frozen at their first observed state until an
incidental refetch. Map parallel.* events to the stages list, run events,
and graph keys.
Also:
- Label branch rows with formatStageLabel so a re-entered branch renders
as `review_glm@2`, matching the sidebar and waterfall.
- Build branch rows in one pass and count live outcomes in one loop.
- Name ParallelBranchId in the OpenAPI spec and reuse fabro_types::
ParallelBranchId, replacing two copies of an inline string format.
- Hoist makeStage and textContent into lib/test-utils so widening Stage
cannot leave per-file fixtures stale (tests are excluded from
typecheck, so the two component-test copies had already gone stale).
- Query stat tiles by data-stat instead of an exact Tailwind class.
- Reuse append_scoped_stage_event's body via append_event_with_scope and
add test_branch_event instead of poking envelope fields.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Splitting on the first colon in FromStr broke bare model IDs that
legitimately contain one. A reference like "llama3:8b" parsed as
provider "llama3" selector "8b", and since "llama3" is not a provider
the lookup failed instead of passing the ID through to the pinned
provider. Verified against origin/main: canonical_session_model with
"future-model:latest" pinned to openrouter returned the passthrough
before and a 400 after.
This is not fixable by choosing a different separator. Bedrock
inference-profile ARNs contain both colons and slashes, and
docs/public/integrations/bedrock.mdx tells users to put arbitrary
inference-profile IDs in api_id. Only the registry can tell a provider
prefix from a model ID that happens to contain the separator.
FromStr now leaves colon-bearing tokens bare, and ModelRef::qualify
promotes only those whose prefix names a known provider. resolve()
applies it, so the fallback path is covered; sessions.rs applies it
before its own match so it keeps its tailored ambiguity messages.
ModelRegistry is now implemented for Catalog in fabro-types, replacing
the CatalogModelRegistry wrapper that existed only in start.rs, so both
call sites share one registry view.
Covered by regression tests at both surfaces, plus qualify unit tests
for ollama tags and Bedrock ARNs. The pre-existing passthrough test
canonical_session_model_preserves_unknown_passthrough_on_selected_provider
passes again.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Both from Copilot review feedback on #652.
Do not fall back to `base_sha` for `final_git_commit_sha`:
`base_sha` is where the run started, not what it produced. When a run made
commits but no SHA was tracked, the conclusion reported the base commit as the
run's final commit — a durable, API-exposed field — and publish then checked
the pushed branch against it, failing a branch that was pushed correctly.
The SHA is now only required where it is actually used: verifying the remote
head before opening a pull request. Pushing never needed it, since the refspec
sends whatever the branch points at. A run with no tracked SHA therefore still
pushes its branch and succeeds; it fails only if a pull request is requested,
where an unverifiable head is a real problem.
Route branch names with slashes in the GitHub twin:
Run branches are `fabro/run/<id>`. GitHub routes the branch as the remainder
of the path, but the twin declared a single-segment `{branch}` capture, so
every real run branch 404'd against it. Now a wildcard, with a test covering
the slashed case that the existing single-segment tests missed.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Conflict in apps/fabro-web/app/components/interview-dock.tsx. Main moved
the dock onto the shared collapsible `RunDockShell` and replaced the
local button constants with shared ones.
Kept main's structure whole and re-applied the review target rendering
onto it: the question paragraph in the shell's `body` becomes the linked
`ReviewTargetQuestion` when the target passes `safeReviewTarget`, and
plain text otherwise. Both now share main's paragraph classes through
`QUESTION_TEXT`, so the two renderings stay visually identical.
`peek` keeps using `question.text`, which is the correct plain-text
collapsed summary for a review target question.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Replace the include_nested flag and its two wrapper functions with a
single outermost-only scanner. Routing extraction used the nested scan
and reverse iteration, so a routing object nested inside a wrapper could
win over its parent -- the same bug class this branch fixes for custom
schemas. No caller needs nested candidates.
Custom schema validation now walks candidates from the end and takes the
last one that parses, instead of parsing only the final candidate. Prose
after the object can contain braces, and outermost-only scanning made
that trailing text a candidate that shadowed the real JSON. Schema errors
are still reported from the last parsable object, so an earlier object
that happens to validate cannot mask a later violation.
Add direct scanner coverage for nesting, adjacent objects, unclosed
braces, and braces inside strings.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
