Post-meeting execution + merge onto main

[pgoel813]

Post-meeting execution (Doc 07) + merge onto main

52 commits, 66 files, +12,272/−15 against main. Implements Doc 07's post-meeting execution
pipeline (B1→B8), lands Doc 04 §112's Workroom dispatch wrapper, and merges main's service
restructure back in.

Read the "Known gaps" section before approving. Several pieces are deliberately unwired,
and one architectural decision is left open for you rather than made here.

---

What this adds

The B1→B8 pipeline (services/control-plane/src/control_plane/post_meeting/):
extract → triage → clarify → plan → approval gate → dispatch → report → final gate.
Tiering is model judgment (Law 4); code owns only the §3.1 floor — if the model claims the
draft tier while reporting an unmet condition, code drops a tier. Nothing ever raises a tier
the model did not ask for.

Four migrations, 0010–0013: post_meeting_tasks, clarify_items, the
staged_drafts.status CHECK constraint, and post_meeting_tasks.planned_at.

Doc 04 §112: the dispatch_workroom tool wrapper, the completion callback, and the
post-meeting completion sink.

Three seams: close→intake (SEAM 1), plan approval (SEAM 2), accept/request-changes
(SEAM 3).

The merge

main dissolved the services/harness member and cut the live path over to
services/in-meeting, deleting the RunLoop spine and the old live brain. All 18 relocations
were taken from git's own rename detection rather than moved by hand — each came back as
CONFLICT (file location) with a suggested destination and no content conflict.

Four judgement calls, each recorded in the merge commit:

• config/defaults.toml — both sections kept. The rejoin_* bare keys are placed
  before the [post_meeting] header; after it they would have been silently reparented
  into our section. Verified by parsing the result.
• live_brain.py — resolved as a delete. Git's modify/delete default left our version
  in the tree; keeping it would have resurrected a module retired deliberately.
• app.py — hand-merged last, after the package underneath was in place.
• The live dispatch path is a dead end — see below.

Known gaps — please read

1. The live in-meeting dispatch path is dead, and this PR does not decide its replacement.

Our §112 work re-woke Proxy by putting the terminal Envelope on run_loop.queue. There is
no run loop. live_sink and the tool mount move, unwired and imported by nothing, to
control_plane/live_dispatch_deadend.py; their nine tests skip at module level.

in_meeting.trigger.EngagementTrigger.on_worker_done is the structural replacement seam — the
same concept, different payload. But choosing between our durable operation_runs claim
(recoverable, cross-process exclusion via the partial unique index) and the warm in-meeting
sandbox toolbelt (faster, no run row, no reaper story) is a founder decision. Porting the
code would have made that choice silently. SessionDriver has no production caller on either
side.

2. The plan-approval route is still not mounted. The blocker changed: it used to be that
§112 did not exist, and every approval produced a permanently orphaned task. §112 now exists
and make_plan_dispatcher supplies a real dispatch=. What is missing is proof — the
end-to-end tests through the route against real Postgres were never written. The mount is one
line and goes in once they are green.

3. estimate_cost has no production supplier. run_dispatch(estimate_cost=None) skips
the §3.5 pre-dispatch cost gate entirely, so task_cost_ceiling is never consulted and
AC-PME-12 does not fire in production. Same shape as the SEAM 1 defect fixed in this PR.
Reported, not fixed — flagging rather than expanding scope.

4. The entire B7 report block has no production caller. build_report,
select_channel, build_draft_card, and deliver are reachable only from tests.

The defect class this PR is mostly about

Three separate wires in this branch looked connected, were exercised only by tests that
supplied the missing piece themselves, and did nothing in production. The last one is fixed
here:

SEAM 1 was dead. The call site had existed since B1 — scribe_runtime calls
_run_post_meeting_intake right after run_close_pass returns. Nothing ever set
CloseConfig.post_meeting_intake. So on every production close the hook was None, the seam
returned at its first line, and run_extract / run_triage / run_clarify / run_plan had
no production caller at all. Every action item in every meeting silently failed to become
a task. test_seam1_close_intake.py passed throughout — every one of its cases constructs the
hook itself.

Fixed three ways: a real supplier (make_intake_hook(db), wired at the only production
construction site, resolving tenant server-side from the meeting row); a loud ERROR on a
missing hook naming the consequence and the fix, replacing a silent return; and a test that
injects nothing and drives the real chain into real Postgres.

That test runs the production builder in a subprocess, which is load-bearing. settings
parses env once at import and caches it, so importing the real boot module in-process under
the fake env its gate demands poisons that cache. The first draft did exactly that and flipped
two genuinely failing doc04 boot tests to passing — doc07 + doc04 reported 529 passed while
doc04 alone reported 2 failed.

Migrations

Both branches independently added a 0009 whose down_revision was
0008_substrate_schema_gaps, giving 0008 two children and alembic two heads. Git cannot
see this — the filenames differ, so the merge succeeds silently and the breakage surfaces
later as alembic upgrade head refusing to run. A double head is not a merge conflict; it is
a clean merge that produces an un-migratable tree.

Ours re-parents onto 0009_repo_maps and renumbers to 0010–0013. Verified in the merged
tree via the CLI: alembic heads → one; upgrade head clean on a fresh database, with both
branches' tables coexisting and each migration's effects present, not just the version stamp.

test_b8_final_gate now locates migrations by suffix glob rather than by number — pinning
the digits meant a pure re-parenting broke a test that has nothing to do with ordering.

Test results

Every suite, each side against its own schema. (An earlier sweep of mine ran main
against a merged-schema database and reported 31 phantom substrate failures; these numbers are
the corrected ones.)

suite	main	this branch	drift
doc00	14F / 197P	14F / 197P	none
doc01	3F / 13P	3F / 13P	none
doc02	11F / 241P	11F / 241P	none
doc03	1F / 412P	1F / 412P	none
doc04	9F / 163P	9F / 188P	same 9 ids, +25 passing
doc05	8F / 305P	8F / 305P	none
doc07	absent	269 passed	entirely new
doc08	4F / 247P	4F / 246P	none (+2 flaky)
e2e	6F / 26P	6F / 26P	none
eval	2F / 22P	2F / 22P	none
code_intel	18F / 2P	18F / 2P	none
hooks / reality	7P / 2P	7P / 2P	none
scripts	4F / 23P	4F / 23P	none
security	11P	11P	none

Zero drift attributable to this branch. doc07 is 269/269 against real Postgres.

The 9 doc04 failures are inherited, not introduced — the same 9 test ids fail on
untouched main. All 9 fail on MeetingRuntime.run_loop, an attribute main's delete-wave
removed while leaving the tests referencing it. Worth fixing on main separately.

test_connect_page.py in doc08 is flaky: two consecutive runs of the identical tree gave 4F
and 6F. The stable set is 4, on both branches.

Not touched

docs/07* and acceptance/doc07/ are byte-identical to their pre-merge state. The four
deleted v2-* docs and the doc03 criteria edit come from main, verified absent there rather
than assumed.

🤖 Generated with Claude Code

Summary by CodeRabbit

• New Features

  • Added post-meeting execution workflows for extracting, triaging, clarifying, planning, approving, and dispatching action items.
  • Added named-human approval gates, ownership checks, concurrency limits, cost controls, and plan expiry handling.
  • Added outcome reporting with channel selection and draft-card fallback.
  • Added staged draft validation that prevents repository pushes and pull requests.
  • Added structured AI output handling with validation and cost reporting.

• Documentation

  • Added Doc 06 and Doc 07 specifications, acceptance criteria, dependency manifests, and implementation guidance.

• Bug Fixes

  • Improved close-flow resilience so post-meeting processing failures do not block meeting completion.

[coderabbitai]

📝 Walkthrough

Walkthrough

Doc 07 adds post-meeting execution from close-record intake through triage, clarification, planning, approval, Workroom dispatch, reporting, and staged-draft handling. It adds PostgreSQL persistence, production wiring, acceptance artifacts, and unit and integration tests.

Changes

Post-meeting execution

Layer / File(s)	Summary
Contracts, acceptance, and persistence acceptance/doc07/*, product/v0-spec/*, migrations/versions/*, config/defaults.toml, libs/llm/*	Defines Doc 07 requirements and criteria. Adds task, clarification, staged-draft, and planning persistence. Adds structured Anthropic output support and post-meeting limits.
Intake, triage, clarification, and planning services/control-plane/src/control_plane/post_meeting/{models,extract,triage,clarify,plan,intake,store,config}.py	Extracts action items, assigns tiers, records unresolved ownership, persists clarification items, generates plans, expires stale plans, and returns guarded intake results.
Approval and Workroom dispatch services/control-plane/src/control_plane/{plan_approval_route,dispatch,dispatch_sinks,live_dispatch_deadend}.py, services/control-plane/src/control_plane/post_meeting/{approval,dispatch,wire}.py	Adds named-human approval, concurrency and cost gates, canonical Workroom dispatch, asynchronous completion handling, and production close wiring.
Reporting, drafts, and outcomes services/control-plane/src/control_plane/post_meeting/{report,final_gate,outcome,owner_actions}.py, services/control-plane/src/control_plane/accept_route.py	Classifies and delivers reportable outcomes. Validates proposed drafts without repository writes. Records accept, change-request, reject, downgrade, and plan-edit outcomes.
Integration validation tests/doc07/*, tests/doc04/*, tests/conftest.py	Adds broad unit, seam, dispatch-wrapper, and real-Postgres coverage. The test configuration requires a database or an explicit waiver for store-backed tests.

Estimated code review effort: 5 (Critical) | ~120 minutes

Sequence Diagram(s)

sequenceDiagram
  participant CloseRuntime
  participant IntakeHook
  participant PostMeetingPipeline
  participant PostgreSQL
  participant PlanApprovalRoute
  participant Workroom
  participant ReportDelivery

  CloseRuntime->>IntakeHook: invoke with finalized notes and meeting ID
  IntakeHook->>PostMeetingPipeline: extract, triage, clarify, and plan
  PostMeetingPipeline->>PostgreSQL: persist task lifecycle records
  PlanApprovalRoute->>PostgreSQL: record named-human approval
  PlanApprovalRoute->>Workroom: dispatch approved task
  Workroom-->>ReportDelivery: return completion envelope
  ReportDelivery->>PostgreSQL: record draft and task outcome

Loading

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 48.02% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly identifies the primary post-meeting execution work and the accompanying merge onto main.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches 💡 2

📝 Generate docstrings 💡

• Create stacked PR
• Commit on current branch

🛠️ Fix failing CI checks 💡

• Fix failing CI checks

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch feat/doc04-112-workroom-dispatch

---

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

Actionable comments posted: 65

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@acceptance/doc07/criteria/criteria.yaml`:
- Around line 763-771: Update the schema_gap note in the criteria configuration
to acknowledge that migration 0012_staged_drafts_status_check.py adds the
staged_drafts_status_enum CHECK constraint. Remove the stale recommendation to
add a CHECK constraint, and preserve the distinction between AC-PME-15-NEG and
AC-PME-07-NEG.
- Around line 451-465: Replace obsolete services/harness citations with the
current control-plane locations and update the runtime/entrypoint scan while
preserving the existing blocker: in acceptance/doc07/criteria/criteria.yaml
lines 451-465 and 511-520, acceptance/doc07/dependency_manifest.yaml lines
28-41, and acceptance/doc07/manifest.yaml lines 71-82, reference
services/control-plane/src/control_plane/meeting_runtime.py:53 and
dispatch.py:51 for MeetingRuntime and assemble_bundle, and ensure the evidence
reflects the current absence of a no-media worker entrypoint.

In `@acceptance/doc07/dependency_manifest.yaml`:
- Around line 28-41: Update the blocked_dependencies entry for the missing
meeting_runtime worker to reflect the current tree: replace the stale
services/harness ownership and
services/harness/src/harness/meeting_runtime.py:54 evidence with a valid
services/in-meeting path and resolved file:line citation, or state “not found by
this method” when no current citation can be established. Apply the same
correction to the corresponding entry in criteria.yaml.

In `@acceptance/doc07/manifest.yaml`:
- Around line 28-38: Update the manifest’s mandatory_faults count from 58 to 60,
matching the 60 unique fault_model_refs declared by criteria.yaml while leaving
the other counts unchanged.

In `@docs/gaps/DOC04-WORKROOM-DISPATCH-UNWIRED.md`:
- Line 56: Update the fenced code block around the ASCII chain diagram in
DOC04-WORKROOM-DISPATCH-UNWIRED.md to specify the text language, using a text
fence while preserving the diagram content.
- Around line 94-96: Update DOC04-WORKROOM-DISPATCH-UNWIRED.md to state that the
approval route records the approval and returns 202 dispatch_blocked with “no
dispatcher is configured” when no dispatcher exists, rather than raising
WorkroomDispatchUnavailable. Replace the dispatch module reference with
services/control-plane/src/control_plane/post_meeting/dispatch.py, and revise
closure steps 1–2 to reference the existing make_dispatch_workroom_server and
make_plan_dispatcher implementations, focusing remaining work on mounting the
server, wiring the dispatcher into the application, and adding end-to-end proof.

In `@libs/llm/src/llm/structured.py`:
- Line 25: Optionally update the import in structured.py to source Awaitable and
Callable from collections.abc instead of typing, while retaining Any and
Protocol from typing; no other changes are required.

In `@migrations/versions/0010_post_meeting_tasks.py`:
- Around line 111-112: Update migrations/versions/0010_post_meeting_tasks.py
lines 111-112 to add a BEFORE UPDATE FOR EACH ROW trigger and function that
assign now() to NEW.updated_at for post_meeting_tasks, and drop both in
downgrade(). Apply the same trigger/function maintenance to
migrations/versions/0011_clarify_items.py lines 63-64 for clarify_items,
including matching cleanup in downgrade().

In `@migrations/versions/0012_staged_drafts_status_check.py`:
- Around line 44-60: Update upgrade() to remove the _ENUM module constant and
replace the f-string in the staged_drafts_status_enum ALTER TABLE statement with
the literal status values directly in the CHECK clause. Preserve the existing
normalization statement and constraint behavior.

In `@migrations/versions/0013_post_meeting_planned_at.py`:
- Around line 51-54: Update the index definition in the migration to match the
sweep query filters: use (tenant_id, planned_at) for tenant-scoped lookups while
retaining the state = 'PLANNED' predicate. Do not include meeting_id, since the
reader does not filter by it.

In `@product/v0-spec/05-WORKROOM.md`:
- Line 311: The amendment notes use stale hard-coded line numbers; replace them
with durable section anchors. In product/v0-spec/05-WORKROOM.md at lines
311-311, update the P8 note to cite §2.5.1, §3.8, and §3.13.6 instead of lines
56, 309, and 368. In product/v0-spec/07-POST-MEETING-EXECUTION.md at lines
172-172, remove line numbers while retaining the existing Doc 05 §2 and §3.8
anchors and the code-sample reference as a section anchor.
- Line 56: Update the staged-draft gate wording around propose_change, including
the matching phrasing at the later occurrence, to distinguish the persisted
staged_drafts row status from the returned envelope status: the row remains
status='proposed', while the return carries status='needs_review' alongside
draft_id. Preserve the existing no-landing and human-approval behavior.

In `@product/v0-spec/06-PROACTIVE.md`:
- Around line 47-53: Add the `text` language tag to the fenced verdict-schema
block in the proactive specification, and apply the same fenced-block language
annotation to the corresponding schema in Doc 07. Preserve the schema contents
unchanged.
- Around line 143-145: Update the clarify_items ownership wording in Doc 06,
including the schema note and the repeated ownership statement, to state that
both Doc 06 and Doc 07 write rows, with Doc 07 also completing them. Preserve
the existing question, kind, blocking_ref, urgency, answer, and answered_by
field details.

In `@product/v0-spec/07-POST-MEETING-EXECUTION.md`:
- Line 134: Update both §4 references, including the D07.1 CANONICAL-DECISIONS
text, to remove the withdrawn “meeting_runtime worker without a media session”
hosting model. State instead that SessionDriver resolves the run from (db,
meeting_id) and that no new deployable is introduced, while preserving the
surrounding execution and durability requirements.

In `@product/v0-spec/AMENDMENTS-06-07.md`:
- Line 5: Update the patch-count references in the document introduction and
checklist summary: change all three occurrences claiming eight patches—including
the line 5 and line 22 statements and the “five of the eight patches” wording
near line 133—to reflect the nine patches defined by P1 through P9. Do not
modify any other content.

In `@product/v0-spec/CANONICAL-DECISIONS.md`:
- Line 224: Update the §2 DDL comment referenced near the `meeting_id` schema
definition to remove the stale `task_id` association and state that `scope_id`
contains the `meeting_id` only. Keep the existing P10 ruling and UUID/type
requirements unchanged, including the documented `meeting_id::text` cast at the
atomic claim call site.

In `@services/control-plane/src/control_plane/accept_route.py`:
- Around line 271-286: Update the docstring in
services/control-plane/src/control_plane/accept_route.py:271-286 to reference
control_plane.post_meeting.outcome instead of harness.post_meeting.outcome. In
migrations/versions/0010_post_meeting_tasks.py:18-20, replace the obsolete
services/harness/src/harness/dispatch.py:129-145 citation with the current
file:line location of the workroom:{task_id} dispatch wrapper, or state not
found by this method if it cannot be located.
- Line 115: Update install_accept_route and install_reject_route to accept
post_meeting_outcome and pass it into their route handlers, then update app.py
to mount both routes with a sink adapter that invokes the configured
meeting-outcome service. Add tests covering both mounted HTTP paths and verify
accepted or rejected tasks are no longer left in DRAFTED.
- Around line 178-184: Update the outcome handling around _post_meeting_outcome
in the accept and decline flows to skip task closure when
applied.already_applied or declined.already_applied is true. Wire the mounted
routes with a synchronous adapter that invokes the appropriate asynchronous
record_accept or record_changes_requested operation and waits for its
completion, rather than passing the coroutine-returning methods directly to
dispatch_sinks.post_meeting_sink.

In `@services/control-plane/src/control_plane/dispatch.py`:
- Around line 227-235: Update the _DISPATCH_TOOL_DESCRIPTION text to describe
dispatching work to run in the background rather than naming the internal
“Workroom” component. Preserve the existing guidance about when to use the tool,
its arguments, asynchronous task_id response, and repository behavior, while
removing all user-visible internal component names.
- Around line 316-317: Update the exception handler around the completion
callback in the dispatch callback flow to catch Exception instead of
BaseException, preserving logging for callback failures such as RuntimeError
while allowing KeyboardInterrupt and SystemExit to propagate.
- Around line 287-291: Move the asyncio and logging imports, along with the log
logger initialization, from run_and_notify to module scope beside the existing
imports. Remove the per-call imports and logger creation from run_and_notify,
while preserving its existing task-scheduling behavior.
- Around line 391-411: The successful cost-gated path in the dispatch flow must
expose the claimed run identifier. Update dispatch_workroom and the surrounding
handling so a successful call returns or resolves a WorkroomHandle containing
run_id before the existing run_id check; preserve the None path for budget
refusals, then continue to run_task and run_and_notify with the resolved
identifier.

In `@services/control-plane/src/control_plane/live_dispatch_deadend.py`:
- Around line 130-135: Update the preserved import in the revival path around
live dispatch to reference make_dispatch_workroom_server from
control_plane.dispatch instead of harness.dispatch, and extend the restoration
checklist near the existing live_sink and mount-point items to record this
required import correction.

In `@services/control-plane/src/control_plane/plan_approval_route.py`:
- Around line 19-23: Update the docstring references near the dispatch wiring
and the corresponding references at the later noted section to use
control_plane.post_meeting.wire.make_plan_dispatcher instead of
harness.post_meeting.wire.make_plan_dispatcher. Keep the described dispatch
behavior unchanged.
- Around line 253-260: Update _write_approved to return the cursor’s
affected-row count after the guarded UPDATE, rather than always returning None.
In handle_approve_plan, treat a zero count as a lost race and return HTTP 409;
only report approved=True and 200 when the update affects a row.
- Around line 229-260: Update _load_task_row and _write_approved to use asyncpg
asynchronously: await fetchrow and execute directly on the pool/connection with
$1-style parameters, and make both helpers async. Update their callers in the
approval route to await them, preserving the existing tenant/state checks and
single-statement approval update.

In `@services/control-plane/src/control_plane/post_meeting/__init__.py`:
- Around line 3-6: Update the module docstring in the post_meeting package to
replace the stale services/harness and harness-hosting references with
services/control-plane and its control-plane host service. Preserve the existing
meeting_runtime, no-new-deployable, and AC-REPO-006 justification while naming
the relocated package’s actual service.

In `@services/control-plane/src/control_plane/post_meeting/clarify.py`:
- Around line 182-189: Update run_clarify and PostMeetingTaskStore.set_state to
accept and use tenant_id, constraining the task-state update WHERE clause to
both tenant_id and task_id. Validate task_id before the existing try block, and
make set_state fail when no row is updated so clarification insertion cannot
proceed for an unmatched task.
- Around line 166-173: Normalize nullable owner and text values in the
item-processing loop of run_clarify before converting them to strings. Map a
None owner to UNRESOLVED and a None text value to an empty string, while
preserving existing non-null values and downstream assess behavior.

In `@services/control-plane/src/control_plane/post_meeting/config.py`:
- Around line 16-38: Update load_post_meeting_config to validate each resolved
limit after type coercion: max_concurrent_tasks, max_tasks_per_meeting,
task_cost_ceiling, and plan_expiry must be positive. For any non-positive value,
use that key’s value from _FALLBACK while preserving valid configured values and
the existing draft_tier_enabled behavior.
- Around line 41-49: Replace the deep libs.db.src.* and libs.llm.src.* imports
across the post-meeting modules with the repository’s canonical db and llm
workspace import seams; in config.py, use db.config.load_defaults, and apply the
corresponding canonical imports in plan.py, triage.py, and wire.py. In wire.py,
import the HTTP facade from libs.http.external, preserving existing behavior and
exposing fallback observability through db.config.load_defaults if required.

In `@services/control-plane/src/control_plane/post_meeting/dispatch.py`:
- Around line 176-208: Prevent failed Workroom dispatches from leaving tasks
stuck in RUNNING: update the flow around assemble_bundle, store.set_state, and
workroom_dispatch so RUNNING is written only after a successful handle with a
non-null run_id, or explicitly revert the task to APPROVED before returning
ERROR from both the dispatch exception and missing-run_id branches. Preserve the
existing error outcomes and ensure no failed claim retains a concurrency slot.

In `@services/control-plane/src/control_plane/post_meeting/extract.py`:
- Around line 176-180: Replace BaseException with Exception in run_extract at
services/control-plane/src/control_plane/post_meeting/extract.py:176-180,
_record at
services/control-plane/src/control_plane/post_meeting/outcome.py:77-83, and
_dispatch at
services/control-plane/src/control_plane/post_meeting/wire.py:158-163. Preserve
the existing error-result and logging behavior while allowing cancellation and
process-shutdown exceptions to propagate.

In `@services/control-plane/src/control_plane/post_meeting/final_gate.py`:
- Around line 17-19: Update migration references in
services/control-plane/src/control_plane/post_meeting/final_gate.py lines 17-19
and its comment at line 36 from 0011 to 0012_staged_drafts_status_check; in
tests/doc07/_support.py lines 26-29, 41, and 150 from 0009 to
0010_post_meeting_tasks; and in tests/doc07/test_b5_approval.py lines 3-6 from
0009's CHECK and trigger to 0010_post_meeting_tasks. Use suffix-based migration
names in all prose.
- Around line 107-124: Add a validated no-push audit record at the
Workroom/provider boundary, capturing the actual operations and token scopes
associated with each dispatch. Update dispatch_sinks.py’s production path and
_observe_chunk/Envelope flow to preserve this evidence, then pass the record
explicitly to run_final_gate and assert_no_repo_writes instead of relying on
empty defaults.

In `@services/control-plane/src/control_plane/post_meeting/intake.py`:
- Around line 116-131: Update the triage-to-run_clarify mapping in the list
comprehension so has_scope and has_done_condition use independent evidence from
extraction or triage rather than the shared Tier.QUESTION check. Preserve
separate owner, scope, and done-condition signals so run_clarify blocks any item
missing either condition and prevents underspecified items from reaching
planning.
- Around line 147-166: The exception handlers in run_intake and
run_intake_guarded must re-raise asyncio.CancelledError before handling other
BaseException values, preserving cancellation propagation to the close task.
Update both “total”/“never raises” docstrings and related comments to explicitly
exclude cancellation while retaining existing handling for non-cancellation
failures.

In `@services/control-plane/src/control_plane/post_meeting/plan.py`:
- Around line 154-175: Update run_plan’s generate_structured error handling to
catch any Exception, not only StructuredOutputError, record the exception in
result.error, log the failed item, and return the result so one task’s failure
does not stop processing remaining items in the batch.
- Around line 207-251: Update expire_stale_plans to require a tenant_id and
validate that every task row belongs to that tenant before calling
store.set_outcome; reject or raise on mismatches so cross-tenant rows cannot be
expired. Preserve the existing state, expiry, skip, and error handling for rows
that pass the tenant check, and use the row’s established tenant identifier
field.

In `@services/control-plane/src/control_plane/post_meeting/report.py`:
- Around line 84-85: Define _STATUS_CEILING independently from
CONFIDENCE_BY_STATUS using explicit per-status maximum values, rather than
copying the mapping. Keep status_rank and build_report unchanged so the
AC-PME-14 assertion compares the computed confidence against an independently
declared ceiling and can detect upward rounding.
- Around line 285-295: Update the send loop around report delivery so failed
sends do not add their key to the caller-provided already_delivered/seen set;
only add the key after a successful send. Preserve card surfacing and error
recording for failures, using a separate tracking set if needed to prevent
duplicate card entries without suppressing future delivery retries.

In `@services/control-plane/src/control_plane/post_meeting/store.py`:
- Around line 116-135: Require tenant scoping for all task reads, including
planned_tasks_for_sweep, get, and task_id_for_draft. Do not allow None to
produce cross-tenant results; require a tenant_id predicate in each query or
fail closed with an explicit runtime guard when the fetched row’s tenant_id
differs from the caller’s tenant. Preserve the existing task and draft lookup
behavior for matching tenants.

In `@services/control-plane/src/control_plane/post_meeting/triage.py`:
- Around line 190-208: Make run_triage honor its “never raises” contract by
moving load_post_meeting_config into the protected error-handling path and
broadening handling around build_prompt to catch malformed configuration or item
data, including TypeError and ValueError. On either failure, return the existing
empty TriageResult so items remain untiered, while preserving the current
successful structured-generation behavior.

In `@services/control-plane/src/control_plane/post_meeting/wire.py`:
- Around line 245-269: Update _dispatch around run_and_notify so task-scheduling
failures after dispatch_workroom claims the run are caught, recorded through the
existing failure mechanism, and the unsubmitted coroutine is closed to avoid an
unawaited-coroutine warning. Preserve normal run_and_notify behavior and return
the handle on successful scheduling.

In `@services/control-plane/src/control_plane/scribe_runtime.py`:
- Line 872: Update the hook assignment near the meeting-intake handling to use
direct access via close_config.post_meeting_intake instead of getattr with a
fallback, preserving the existing hook behavior.
- Around line 861-900: Update both run_intake and run_intake_guarded to re-raise
asyncio.CancelledError before their broad BaseException handlers, preserving
cancellation instead of swallowing it. In _run_post_meeting_intake, wrap the
complete await hook(final_notes, meeting_id=meeting_id) operation in an
asyncio.timeout using the project’s configured intake timeout, so database
acquisition and downstream LLM work are bounded; retain the existing failure
logging for timeout and other non-cancellation errors.

In `@tests/doc04/test_dispatch_tool_wrapper.py`:
- Around line 216-240: The sandbox failure path in
test_faults_return_a_composed_actionable_reason expects “Nothing was started”
even though dispatch_workroom has already claimed an operation_runs row. Update
the production flow around dispatch_workroom and its outer error handling to
release the claim when task startup fails, or accurately distinguish post-claim
failures in the composed reason; then revise this test’s assertion to match the
persisted-state behavior while retaining actionable guidance.
- Around line 101-109: Remove the vacuous getattr assertion in
test_no_annotations_so_readonlyhint_stays_false, or replace it with an assertion
against the tool’s serialized shape that directly verifies readOnlyHint is
false. Keep the existing annotations is None structural check if no serialized
representation is available.
- Around line 254-288: Update
test_the_cost_gate_declining_is_not_an_error_but_is_not_accepted to remove the
unused GatedDB class and use the monkeypatch fixture to replace
mod.dispatch_workroom, eliminating manual restoration. Also update the dispatch
handling to distinguish accepted and declined DispatchDecision results using
their dispatched status rather than inferring from run_id or handle presence.

In `@tests/doc07/_support.py`:
- Around line 295-313: Update the AST scanning logic in the surrounding helper
to use a single ast.walk(tree) pass: retain the existing import, name, and
attribute collection, remove the no-op ast.Constant branch, and handle string
constants from ast.Call arguments inline during that same traversal. Preserve
the current behavior of collecting only call-argument strings.

In `@tests/doc07/test_b2_triage.py`:
- Around line 173-185: Update
test_ac_pme_06_each_violated_condition_drops_a_tier so its parametrization
reflects actual triage behavior: since run_triage uses draft_conditions_met and
ignores failed_conditions, reduce the test to one representative met=False case
and explicitly assert failed_conditions is advisory only, or modify the
triage/verdict flow to record and use failed_conditions before retaining the
per-condition parametrization.

In `@tests/doc07/test_b3_clarify.py`:
- Around line 227-240: Fix
test_ac_pme_04_neg_malformed_channel_set_sends_nothing so it asserts an
observable delivery outcome instead of the never-written sent list: remove the
ineffective sent assertion and verify res.outcomes[0].delivery_deferred is False
and routed_to is None. Keep the existing pending-state and task-state
assertions, and align the test with run_clarify’s current interface since it
accepts no sender.

In `@tests/doc07/test_b6_dispatch.py`:
- Around line 127-129: Update the banned-token check in the test around lowered
source so the “class .*queue” pattern is evaluated as a regular expression via
re.search, or replace it with the intended literal token. Preserve the existing
checks for “sandboxprovider(”, “e2b.”, and “scheduler(”.

In `@tests/doc07/test_b7_report.py`:
- Around line 233-235: Replace the hard-coded source path in
tests/doc07/test_b7_report.py:233-235 by importing
control_plane.post_meeting.report and resolving pathlib.Path(report.__file__).
Apply the same change in tests/doc07/test_seam2_plan_approval.py:227-229, using
the existing control_plane.plan_approval_route module import (m) and
pathlib.Path(m.__file__).

In `@tests/doc07/test_integration_claim_recycle.py`:
- Line 157: Replace suffix-based sweep count assertions around the integration
test’s sweep results with exact status-string comparisons: compare swept to
"UPDATE 1", the zero-row result at the later assertion to "UPDATE 0", and
first/second to "UPDATE 1"/"UPDATE 0" respectively. Preserve the existing
assertion messages and test flow.

In `@tests/doc07/test_integration_db.py`:
- Around line 14-17: Update the test documentation to use asyncpg exception
terminology, including references to asyncpg.exceptions.* and non-integrity
RaiseError. Revise the migration references from 0011 to
0012_staged_drafts_status_check.py and from 0012 to
0013_post_meeting_planned_at.py, while preserving the staged_drafts_status_enum
constraint name.

In `@tests/doc07/test_owner_actions.py`:
- Around line 151-157: Replace the dynamic __import__ expression in the
expire_stale_plans call with a module-level import of PostMeetingConfig,
matching the existing pattern in test_b4_plan.py. Use the imported
PostMeetingConfig directly when constructing the config argument, leaving the
test behavior unchanged.

In `@tests/doc07/test_seam1_close_intake.py`:
- Around line 139-159: Update
test_ac_pme_02_a_raising_intake_never_reaches_the_close and the
_run_post_meeting_intake contract to distinguish cancellation from other
BaseException failures: add asyncio.CancelledError coverage that asserts
cancellation propagates, while retaining the existing expectation that
RuntimeError, ConnectionRefusedError, KeyboardInterrupt, and MemoryError are
swallowed. Ensure the implementation does not absorb task cancellation.
- Around line 219-235: Replace the tautological assertions in
test_ac_pme_02_neg_run_intake_itself_never_raises and
test_ac_pme_02_neg_guarded_wrapper_returns_none_on_total_failure with checks
that garbage or invalid dependencies produce a failed result rather than silent
success, and verify the result identifies the failed intake stage. Preserve the
guarded wrapper’s documented None outcome if applicable, but when it returns a
result assert failure and the corresponding stage name.

In `@tests/doc07/test_seam3_outcome.py`:
- Around line 151-156: Add an asynchronous test case alongside
test_route_guard_passes_the_action_and_actor that injects an async sink into
_post_meeting_outcome, awaits the route helper as required, and asserts the
recorded action, draft_id, and who values are observed. Keep the existing
synchronous-shape coverage intact while ensuring coroutine sinks are awaited.
- Around line 70-92: Update test_all_three_terminal_states_are_now_reachable so
the DISCARDED case is produced through an actual plan-expiry function or B8
final-gate refusal, rather than calling store.set_outcome directly; retain the
assertion that all three terminal states are reachable. If no real path can be
exercised here, instead rename the test and narrow its docstring and expected
states to the two outcomes written by record_accept and
record_changes_requested.
- Line 21: Replace the dynamic __import__("datetime") expression used to define
NOW with a normal typed datetime import, while preserving the existing UTC
datetime value and strict mypy type checking.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: ae2dc963-ff79-4342-a31f-a7d42af51025

📥 Commits

Reviewing files that changed from the base of the PR and between 4b26e8b and dc4298c.

⛔ Files ignored due to path filters (1)

• uv.lock is excluded by !**/*.lock

📒 Files selected for processing (65)

• CLAUDE.md
• acceptance/doc07/criteria/criteria.yaml
• acceptance/doc07/dependency_manifest.yaml
• acceptance/doc07/manifest.yaml
• acceptance/doc07/requirements/requirements.yaml
• config/defaults.toml
• docs/gaps/DOC04-WORKROOM-DISPATCH-UNWIRED.md
• libs/llm/pyproject.toml
• libs/llm/src/llm/structured.py
• migrations/versions/0010_post_meeting_tasks.py
• migrations/versions/0011_clarify_items.py
• migrations/versions/0012_staged_drafts_status_check.py
• migrations/versions/0013_post_meeting_planned_at.py
• product/v0-spec/05-WORKROOM.md
• product/v0-spec/06-PROACTIVE.md
• product/v0-spec/07-POST-MEETING-EXECUTION.md
• product/v0-spec/AMENDMENTS-06-07.md
• product/v0-spec/CANONICAL-DECISIONS.md
• product/v0-spec/PLATFORM-ADOPTION.md
• product/v0-spec/SPINE-REGISTER.md
• services/control-plane/src/control_plane/accept_route.py
• services/control-plane/src/control_plane/app.py
• services/control-plane/src/control_plane/dispatch.py
• services/control-plane/src/control_plane/dispatch_sinks.py
• services/control-plane/src/control_plane/live_dispatch_deadend.py
• services/control-plane/src/control_plane/plan_approval_route.py
• services/control-plane/src/control_plane/post_meeting/__init__.py
• services/control-plane/src/control_plane/post_meeting/approval.py
• services/control-plane/src/control_plane/post_meeting/clarify.py
• services/control-plane/src/control_plane/post_meeting/config.py
• services/control-plane/src/control_plane/post_meeting/dispatch.py
• services/control-plane/src/control_plane/post_meeting/extract.py
• services/control-plane/src/control_plane/post_meeting/final_gate.py
• services/control-plane/src/control_plane/post_meeting/intake.py
• services/control-plane/src/control_plane/post_meeting/models.py
• services/control-plane/src/control_plane/post_meeting/outcome.py
• services/control-plane/src/control_plane/post_meeting/owner_actions.py
• services/control-plane/src/control_plane/post_meeting/plan.py
• services/control-plane/src/control_plane/post_meeting/report.py
• services/control-plane/src/control_plane/post_meeting/store.py
• services/control-plane/src/control_plane/post_meeting/triage.py
• services/control-plane/src/control_plane/post_meeting/wire.py
• services/control-plane/src/control_plane/scribe_runtime.py
• services/control-plane/src/control_plane/server.py
• tests/conftest.py
• tests/doc02/test_m1_invite_launches_real_bot.py
• tests/doc04/test_dispatch_tool_wrapper.py
• tests/doc04/test_live_dispatch_deadend.py
• tests/doc07/_support.py
• tests/doc07/test_b1_extract.py
• tests/doc07/test_b2_triage.py
• tests/doc07/test_b3_clarify.py
• tests/doc07/test_b4_plan.py
• tests/doc07/test_b5_approval.py
• tests/doc07/test_b6_dispatch.py
• tests/doc07/test_b7_report.py
• tests/doc07/test_b8_final_gate.py
• tests/doc07/test_integration_claim_recycle.py
• tests/doc07/test_integration_db.py
• tests/doc07/test_owner_actions.py
• tests/doc07/test_seam1_close_intake.py
• tests/doc07/test_seam1_production_wiring.py
• tests/doc07/test_seam2_plan_approval.py
• tests/doc07/test_seam3_outcome.py
• tools/linux-verify-requirements.txt

[coderabbitai]

📐 Maintainability & Code Quality | 🟠 Major | ⚡ Quick win

🧩 Analysis chain

🏁 Script executed:

#!/bin/bash
# Confirm whether services/harness still exists on this branch.
fd -t d 'harness' services || echo 'no services/harness directory'
fd -t f 'meeting_runtime.py' services
# Find the relocated dispatch/assemble_bundle definition.
rg -n --type=py 'def assemble_bundle'
# List every stale citation in the acceptance bundle.
rg -n 'services/harness|harness\.dispatch|harness/dispatch' acceptance product

Repository: dakshparikh12/proxy

Length of output: 217

---

🏁 Script executed:

#!/bin/bash
set -u
printf '%s\n' '--- relevant files ---'
git ls-files \
  acceptance/doc07/criteria/criteria.yaml \
  acceptance/doc07/dependency_manifest.yaml \
  acceptance/doc07/manifest.yaml \
  services/control-plane/src/control_plane/meeting_runtime.py \
  services/in-meeting \
  services/harness

printf '%s\n' '--- current service directories ---'
fd -t d -d 3 . services | sort | head -200

printf '%s\n' '--- runtime and dispatch symbols ---'
rg -n --glob '*.py' 'class MeetingRuntime|def main\(|if __name__|def assemble_bundle|assemble_bundle|harness\.dispatch|services/harness|services/in-meeting|services/control-plane' services acceptance/doc07

printf '%s\n' '--- acceptance references with context ---'
rg -n -C 3 'services/harness|harness\.dispatch|harness/dispatch|assemble_bundle|meeting_runtime' acceptance/doc07

Repository: dakshparikh12/proxy

Length of output: 16840

---

🏁 Script executed:

#!/bin/bash
set -u
for f in \
  acceptance/doc07/criteria/criteria.yaml \
  acceptance/doc07/dependency_manifest.yaml \
  acceptance/doc07/manifest.yaml \
  services/control-plane/src/control_plane/meeting_runtime.py
do
  if [ -f "$f" ]; then
    printf '\n--- %s ---\n' "$f"
    wc -l "$f"
    case "$f" in
      acceptance/doc07/criteria/criteria.yaml) sed -n '440,565p' "$f" ;;
      acceptance/doc07/dependency_manifest.yaml) sed -n '20,78p' "$f" ;;
      acceptance/doc07/manifest.yaml) sed -n '63,88p' "$f" ;;
      services/control-plane/src/control_plane/meeting_runtime.py) sed -n '1,130p' "$f" ;;
    esac
  else
    printf 'not found: %s\n' "$f"
  fi
done

Repository: dakshparikh12/proxy

Length of output: 18891

---

🏁 Script executed:

#!/bin/bash
set -u
printf '%s\n' '--- MeetingRuntime lifecycle and worker entrypoints ---'
ast-grep outline services/control-plane/src/control_plane/meeting_runtime.py
rg -n -C 4 'def __post_init__|def start\(|def aclose\(|carrier|_hearing|start_meeting_scribe|consent_gate|stt_refresh' services/control-plane/src/control_plane/meeting_runtime.py
printf '%s\n' '--- main entrypoints in current service source ---'
rg -n --glob '*.py' '(^|[[:space:]])def main\(|if __name__\s*==\s*["'\'']__main__["'\'']' services
printf '%s\n' '--- current dispatch definitions and surrounding references ---'
sed -n '40,80p' services/control-plane/src/control_plane/dispatch.py
sed -n '118,155p' services/control-plane/src/control_plane/dispatch.py
printf '%s\n' '--- all stale acceptance citations, including inherited negative criteria ---'
rg -n -C 1 'services/harness|harness\.dispatch|harness/dispatch' acceptance/doc07

Repository: dakshparikh12/proxy

Length of output: 29003

---

Replace obsolete Doc 07 path citations. The current clone has no services/harness directory. MeetingRuntime is at services/control-plane/src/control_plane/meeting_runtime.py:53, and assemble_bundle is at services/control-plane/src/control_plane/dispatch.py:51.

Update all affected references in acceptance/doc07/criteria/criteria.yaml, acceptance/doc07/dependency_manifest.yaml, and acceptance/doc07/manifest.yaml. Retain the blocker, but ground it in the current runtime and entrypoint scan.

📍 Affects 3 files

• acceptance/doc07/criteria/criteria.yaml#L451-L465 (this comment)
• acceptance/doc07/criteria/criteria.yaml#L511-L520
• acceptance/doc07/dependency_manifest.yaml#L28-L41
• acceptance/doc07/manifest.yaml#L71-L82

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@acceptance/doc07/criteria/criteria.yaml` around lines 451 - 465, Replace
obsolete services/harness citations with the current control-plane locations and
update the runtime/entrypoint scan while preserving the existing blocker: in
acceptance/doc07/criteria/criteria.yaml lines 451-465 and 511-520,
acceptance/doc07/dependency_manifest.yaml lines 28-41, and
acceptance/doc07/manifest.yaml lines 71-82, reference
services/control-plane/src/control_plane/meeting_runtime.py:53 and
dispatch.py:51 for MeetingRuntime and assemble_bundle, and ensure the evidence
reflects the current absence of a no-media worker entrypoint.

Source: Coding guidelines

[coderabbitai]

📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

🧩 Analysis chain

🏁 Script executed:

#!/bin/bash
# Confirm the staged_drafts status CHECK constraint landed in migration 0012.
fd -t f '0012_staged_drafts_status_check.py' migrations | xargs -r cat -n
# Show the original staged_drafts definition the note cites.
fd -t f '0001_substrate.py' migrations --exec rg -n -C4 'staged_drafts|status'

Repository: dakshparikh12/proxy

Length of output: 6459

---

🏁 Script executed:

#!/bin/bash
set -eu
criteria="$(fd -t f 'criteria.yaml' acceptance | head -n 1)"
printf '%s\n' "== $criteria: schema_gap note =="
sed -n '755,775p' "$criteria"
printf '%s\n' "== $criteria: AC-PME-07-NEG =="
sed -n '1098,1115p' "$criteria"
printf '%s\n' "== $criteria: AC-PME-15-NEG =="
sed -n '1362,1380p' "$criteria"
printf '%s\n' "== references to staged_drafts_status_enum and schema_gap =="
rg -n -C2 'staged_drafts_status_enum|schema_gap|AC-PME-07-NEG|AC-PME-15-NEG' "$criteria"

Repository: dakshparikh12/proxy

Length of output: 3364

---

🏁 Script executed:

#!/bin/bash
set -eu
criteria='acceptance/doc07/criteria/criteria.yaml'
printf '%s\n' "== $criteria: schema_gap note =="
sed -n '755,775p' "$criteria"
printf '%s\n' "== $criteria: AC-PME-07-NEG =="
sed -n '1098,1115p' "$criteria"
printf '%s\n' "== $criteria: AC-PME-15-NEG =="
sed -n '1362,1380p' "$criteria"
printf '%s\n' "== relevant references =="
rg -n -C2 'staged_drafts_status_enum|schema_gap|AC-PME-07-NEG|AC-PME-15-NEG' "$criteria"

Repository: dakshparikh12/proxy

Length of output: 3696

---

Update the schema_gap note. Migration 0012_staged_drafts_status_check.py adds the staged_drafts_status_enum CHECK constraint. AC-PME-15-NEG asserts this behavior; AC-PME-07-NEG concerns a different database constraint. Remove the stale follow-up recommendation.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@acceptance/doc07/criteria/criteria.yaml` around lines 763 - 771, Update the
schema_gap note in the criteria configuration to acknowledge that migration
0012_staged_drafts_status_check.py adds the staged_drafts_status_enum CHECK
constraint. Remove the stale recommendation to add a CHECK constraint, and
preserve the distinction between AC-PME-15-NEG and AC-PME-07-NEG.

[coderabbitai]

📐 Maintainability & Code Quality | 🟡 Minor | ⚡ Quick win

Re-ground the blocked-dependency evidence on the current tree.

This entry cites services/harness/src/harness/meeting_runtime.py:54 and dependency_owner: services/harness. This PR removes the harness member and relocates the service to services/in-meeting. A file:line citation that no longer resolves violates the grounded-or-silent rule. Re-cite the current path, or state not found by this method. The same stale citations appear in acceptance/doc07/criteria/criteria.yaml; see the consolidated comment.

As per coding guidelines: "Ground claims in the current clone with file:line, or state not found by this method."

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@acceptance/doc07/dependency_manifest.yaml` around lines 28 - 41, Update the
blocked_dependencies entry for the missing meeting_runtime worker to reflect the
current tree: replace the stale services/harness ownership and
services/harness/src/harness/meeting_runtime.py:54 evidence with a valid
services/in-meeting path and resolved file:line citation, or state “not found by
this method” when no current citation can be established. Apply the same
correction to the corresponding entry in criteria.yaml.

Source: Coding guidelines

[coderabbitai]

🗄️ Data Integrity & Integration | 🟡 Minor | ⚡ Quick win

🧩 Analysis chain

🏁 Script executed:

#!/bin/bash
# Count fault_model_refs entries and criteria in the Doc 07 bundle.
python3 - <<'PY'
import yaml, pathlib
p = pathlib.Path("acceptance/doc07/criteria/criteria.yaml")
docs = yaml.safe_load(p.read_text(encoding="utf-8"))
total = sum(len(c.get("fault_model_refs") or []) for c in docs)
print("criteria:", len(docs))
print("fault_model_refs total:", total)
print("positives:", sum(1 for c in docs if not c["criterion_id"].endswith("-NEG")))
print("negatives:", sum(1 for c in docs if c["criterion_id"].endswith("-NEG")))
PY

Repository: dakshparikh12/proxy

Length of output: 274

---

🏁 Script executed:

#!/bin/bash
set -eu
printf '%s\n' '--- criteria file ---'
nl -ba acceptance/doc07/criteria/criteria.yaml | sed -n '1,240p'
printf '%s\n' '--- manifest counts ---'
nl -ba acceptance/doc07/manifest.yaml | sed -n '20,45p'

Repository: dakshparikh12/proxy

Length of output: 275

---

🏁 Script executed:

#!/bin/bash
set -eu
printf '%s\n' '--- criteria file ---'
sed -n '1,260p' acceptance/doc07/criteria/criteria.yaml
printf '%s\n' '--- manifest counts ---'
sed -n '20,45p' acceptance/doc07/manifest.yaml

Repository: dakshparikh12/proxy

Length of output: 13172

---

🏁 Script executed:

#!/bin/bash
set -eu
python3 - <<'PY'
from pathlib import Path

path = Path("acceptance/doc07/criteria/criteria.yaml")
criterion = None
counts = {}
total = 0

for line in path.read_text(encoding="utf-8").splitlines():
    if line.startswith("- criterion_id: "):
        criterion = line.split(": ", 1)[1].strip()
        counts[criterion] = 0
    elif line.startswith("  - F-"):
        if criterion is None:
            raise SystemExit("fault_model_refs entry has no criterion")
        counts[criterion] += 1
        total += 1

positive = [key for key in counts if not key.endswith("-NEG")]
negative = [key for key in counts if key.endswith("-NEG")]

print("criteria:", len(counts))
print("positive criteria:", len(positive))
print("negative criteria:", len(negative))
print("fault_model_refs total:", total)
print("positive fault_model_refs:", sum(counts[key] for key in positive))
print("negative fault_model_refs:", sum(counts[key] for key in negative))
print("criteria with counts other than 2:", [
    (key, counts[key]) for key in counts if counts[key] != 2
])
PY

Repository: dakshparikh12/proxy

Length of output: 373

---

🏁 Script executed:

#!/bin/bash
set -eu
python3 - <<'PY'
from pathlib import Path

path = Path("acceptance/doc07/criteria/criteria.yaml")
criterion = None
refs = {}
for line in path.read_text(encoding="utf-8").splitlines():
    if line.startswith("- criterion_id: "):
        criterion = line.split(": ", 1)[1].strip()
        refs[criterion] = []
    elif line.startswith("  - F-"):
        refs[criterion].append(line.strip()[2:])

all_refs = [ref for values in refs.values() for ref in values]
unique_refs = sorted(set(all_refs))
duplicates = sorted({ref for ref in all_refs if all_refs.count(ref) > 1})

print("reference entries:", len(all_refs))
print("unique fault IDs:", len(unique_refs))
print("duplicate fault IDs:", duplicates)
for criterion_id in ("AC-PME-07", "AC-PME-13"):
    print(criterion_id, len(refs[criterion_id]), refs[criterion_id])
PY

printf '%s\n' '--- AC-PME-07 and AC-PME-13 context ---'
grep -n -A85 -B5 -E '^[-] criterion_id: AC-PME-07$|^[-] criterion_id: AC-PME-13$' acceptance/doc07/criteria/criteria.yaml

Repository: dakshparikh12/proxy

Length of output: 8245

---

Set mandatory_faults to 60. criteria.yaml contains 60 unique fault_model_refs: 32 positive and 28 negative criteria. The manifest value 58 is incorrect.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@acceptance/doc07/manifest.yaml` around lines 28 - 38, Update the manifest’s
mandatory_faults count from 58 to 60, matching the 60 unique fault_model_refs
declared by criteria.yaml while leaving the other counts unchanged.

[coderabbitai]

📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

🧩 Analysis chain

🌐 Web query:

Does ruff rule UP035 flag importing Callable and Awaitable from typing instead of collections.abc?

💡 Result:

Yes, Ruff rule UP035 (deprecated-import) flags imports of Callable and Awaitable from the typing module when they should be imported from collections.abc [1][2]. This rule is designed to enforce the use of modern, standard-library locations for various classes and types, as dictated by Python's evolution (such as PEP 585) [1][3]. Because these changes can have implications for code running on older Python versions, Ruff's behavior is version-aware: 1. Version Sensitivity: The rule respects the --target-version configuration [4][1]. For example, the rewrite for typing.Callable to collections.abc.Callable was historically excluded for Python 3.9.0 and 3.9.1 due to bugs in those specific versions [2][5]. 2. Scope: UP035 specifically flags the imports themselves (e.g., from typing import Callable) [6][7]. It does not necessarily flag usages of the fully qualified name (e.g., typing.Callable) directly in the code, as the rule is currently implemented to target import statements [6][8]. 3. Relationship with Other Rules: While UP035 handles these imports, other rules like UP006 (non-pep585-annotation) handle the replacement of these types in type annotations (e.g., changing List[int] to list[int]) [1][8]. Due to ongoing development, there can occasionally be overlaps or inconsistencies in how these rules are applied across different Python versions [8]. In summary, if you are targeting a modern Python version, Ruff will suggest moving Callable and Awaitable from typing to collections.abc via UP035 [1][5].

Citations:

• 1: UP035 vs. flake8-pep585 astral-sh/ruff#3388
• 2: UP035: Rewrite typing.Callable to collections.abc.Callable for Python 3.9.2+ astral-sh/ruff#2690
• 3: https://docs.astral.sh/ruff/rules/?featured_on=talkpython
• 4: https://docs.astral.sh/ruff/rules/deprecated-import/
• 5: UP035 vs. flake8-pep585 astral-sh/ruff#3388
• 6: UP035: No violations when importing an entire module and using deprecated/moved class/method astral-sh/ruff#7526
• 7: https://docs.astral.sh/ruff/tutorial/
• 8: Duplication and inconsistency between pep585 annotation rules: UP006, UP035 astral-sh/ruff#15246

---

🏁 Script executed:

#!/bin/bash
set -eu
printf '%s\n' '--- candidate files ---'
fd -t f 'structured\.py|pyproject\.toml|ruff\.toml|\.ruff\.toml' . | head -80
printf '%s\n' '--- import and package metadata ---'
rg -n -C 3 'from typing import Any, Awaitable, Callable, Protocol|requires-python|target-version|select|extend-select|UP035' \
  --glob 'structured.py' --glob 'pyproject.toml' --glob 'ruff.toml' --glob '.ruff.toml' .
printf '%s\n' '--- ruff availability ---'
if command -v ruff >/dev/null 2>&1; then
  ruff --version
else
  echo 'ruff: not installed'
fi

Repository: dakshparikh12/proxy

Length of output: 8090

---

🏁 Script executed:

#!/bin/bash
set -eu
printf '%s\n' '--- configured Ruff result ---'
ruff check libs/llm/src/llm/structured.py
printf '%s\n' '--- explicit UP035 result ---'
ruff check --select UP035 libs/llm/src/llm/structured.py
printf '%s\n' '--- rule explanation ---'
ruff rule UP035 | sed -n '1,100p'

Repository: dakshparikh12/proxy

Length of output: 1383

---

Use collections.abc for Awaitable and Callable as an optional cleanup.

UP035 is not enabled in the configured Ruff rule set, so this change is not required for the current lint configuration.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@libs/llm/src/llm/structured.py` at line 25, Optionally update the import in
structured.py to source Awaitable and Callable from collections.abc instead of
typing, while retaining Any and Protocol from typing; no other changes are
required.

Source: Coding guidelines

[coderabbitai]

🩺 Stability & Availability | 🟡 Minor | ⚡ Quick win

Add asyncio.CancelledError to the parametrization, or assert that it propagates.

The test pins that _run_post_meeting_intake swallows any BaseException. asyncio.CancelledError is a BaseException. If the close guard catches it too, a cancelled close task absorbs its own cancellation, the task keeps running past shutdown, and the event loop cannot stop it cleanly. The current cases (KeyboardInterrupt, MemoryError) do not distinguish the two behaviors.

Decide the contract and pin it. Cancellation should normally propagate while other BaseException classes are contained.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tests/doc07/test_seam1_close_intake.py` around lines 139 - 159, Update
test_ac_pme_02_a_raising_intake_never_reaches_the_close and the
_run_post_meeting_intake contract to distinguish cancellation from other
BaseException failures: add asyncio.CancelledError coverage that asserts
cancellation propagates, while retaining the existing expectation that
RuntimeError, ConnectionRefusedError, KeyboardInterrupt, and MemoryError are
swallowed. Ensure the implementation does not absorb task cancellation.

[coderabbitai]

📐 Maintainability & Code Quality | 🟡 Minor | ⚡ Quick win

Both assertions are tautological and cannot fail.

TriageResult.ok-style properties return self.error is None, so isinstance(res.ok, bool) at line 226 is true for every possible outcome, including a silent success on garbage input. Line 235 is worse: out is None or out.ok in (True, False) holds for any value run_intake_guarded can return. Neither test would catch a regression.

Assert the intended contract: the call returns a failed result, and it names the stage that failed.

Proposed assertions

     res = await run_intake(
         object(), meeting_id=MEETING, tenant_id=TENANT,
         task_store=None, clarify_store=None, caller=None, call_external=None,
     )
-    assert isinstance(res.ok, bool)
+    assert res.ok is False, "garbage input produced a successful intake"
+    assert res.error is not None
+    assert res.failed_stage == "extract"

     out = await run_intake_guarded(
         notes(), meeting_id=MEETING, tenant_id=TENANT,
         task_store=object(), clarify_store=object(), caller=None, call_external=None,
     )
-    assert out is None or out.ok in (True, False)
+    assert out is None or out.ok is False

📝 Committable suggestion

‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.

Suggested change

	@pytest.mark.negative
	async def test_ac_pme_02_neg_run_intake_itself_never_raises():
	"""Even handed garbage, intake returns a result rather than propagating."""
	res = await run_intake(
	object(), meeting_id=MEETING, tenant_id=TENANT,
	task_store=None, clarify_store=None, caller=None, call_external=None,
	)
	assert isinstance(res.ok, bool)
	@pytest.mark.negative
	async def test_ac_pme_02_neg_guarded_wrapper_returns_none_on_total_failure():
	out = await run_intake_guarded(
	notes(), meeting_id=MEETING, tenant_id=TENANT,
	task_store=object(), clarify_store=object(), caller=None, call_external=None,
	)
	assert out is None or out.ok in (True, False)
	`@pytest.mark.negative`
	async def test_ac_pme_02_neg_run_intake_itself_never_raises():
	"""Even handed garbage, intake returns a result rather than propagating."""
	res = await run_intake(
	object(), meeting_id=MEETING, tenant_id=TENANT,
	task_store=None, clarify_store=None, caller=None, call_external=None,
	)
	assert res.ok is False, "garbage input produced a successful intake"
	assert res.error is not None
	assert res.failed_stage == "extract"
	`@pytest.mark.negative`
	async def test_ac_pme_02_neg_guarded_wrapper_returns_none_on_total_failure():
	out = await run_intake_guarded(
	notes(), meeting_id=MEETING, tenant_id=TENANT,
	task_store=object(), clarify_store=object(), caller=None, call_external=None,
	)
	assert out is None or out.ok is False

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tests/doc07/test_seam1_close_intake.py` around lines 219 - 235, Replace the
tautological assertions in test_ac_pme_02_neg_run_intake_itself_never_raises and
test_ac_pme_02_neg_guarded_wrapper_returns_none_on_total_failure with checks
that garbage or invalid dependencies produce a failed result rather than silent
success, and verify the result identifies the failed intake stage. Preserve the
guarded wrapper’s documented None outcome if applicable, but when it returns a
result assert failure and the corresponding stage name.

[coderabbitai]

📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

Use a normal import for datetime.

__import__("datetime") returns Any, which removes type checking from NOW under mypy --strict, and it is harder to read than a module import.

Proposed change

+from datetime import datetime, timezone
+
...
-NOW = __import__("datetime").datetime(2026, 7, 28, tzinfo=__import__("datetime").timezone.utc)
+NOW = datetime(2026, 7, 28, tzinfo=timezone.utc)

📝 Committable suggestion

‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.

Suggested change

	NOW = __import__("datetime").datetime(2026, 7, 28, tzinfo=__import__("datetime").timezone.utc)
	from datetime import datetime, timezone
	NOW = datetime(2026, 7, 28, tzinfo=timezone.utc)

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tests/doc07/test_seam3_outcome.py` at line 21, Replace the dynamic
__import__("datetime") expression used to define NOW with a normal typed
datetime import, while preserving the existing UTC datetime value and strict
mypy type checking.

[coderabbitai]

📐 Maintainability & Code Quality | 🟡 Minor | ⚡ Quick win

The test does not show that DISCARDED is reachable by a real path.

The docstring says "each written by a real path". Lines 84-87 write DISCARDED by calling store.set_outcome directly from the test. That proves the store accepts the value, not that plan expiry or the final gate ever produces it. The comment at line 83 names the two real writers, so the assertion at lines 88-92 reads as coverage that does not exist.

Either drive DISCARDED through the plan-expiry function or the B8 gate, or narrow the docstring and the test name to the two states this test actually exercises.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tests/doc07/test_seam3_outcome.py` around lines 70 - 92, Update
test_all_three_terminal_states_are_now_reachable so the DISCARDED case is
produced through an actual plan-expiry function or B8 final-gate refusal, rather
than calling store.set_outcome directly; retain the assertion that all three
terminal states are reachable. If no real path can be exercised here, instead
rename the test and narrow its docstring and expected states to the two outcomes
written by record_accept and record_changes_requested.

[coderabbitai]

🗄️ Data Integrity & Integration | 🟡 Minor | ⚡ Quick win

This test locks in a synchronous sink shape.

The double is lambda **kw: seen.append(kw), which returns immediately. The real recorders record_accept and record_changes_requested are coroutine functions; lines 44 and 55 await them. This test therefore cannot detect the case where the injected sink is a coroutine function and _post_meeting_outcome never awaits it. See the comment on services/control-plane/src/control_plane/accept_route.py lines 178-184 for the root cause.

Add a case that passes a coroutine function as the sink and asserts the write-back is observed.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tests/doc07/test_seam3_outcome.py` around lines 151 - 156, Add an
asynchronous test case alongside test_route_guard_passes_the_action_and_actor
that injects an async sink into _post_meeting_outcome, awaits the route helper
as required, and asserts the recorded action, draft_id, and who values are
observed. Keep the existing synchronous-shape coverage intact while ensuring
coroutine sinks are awaited.

[coderabbitai]

Review continued from previous batch...

[coderabbitai]

📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

Add a language to this fenced block.

markdownlint reports MD040 for line 56. The block is an ASCII chain diagram, so text is enough.

📝 Proposed fix

-```
+```text
 wake turn (model emits the tool)

🧰 Tools

🪛 markdownlint-cli2 (0.23.1)

[warning] 56-56: Fenced code blocks should have a language specified

(MD040, fenced-code-language)

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@docs/gaps/DOC04-WORKROOM-DISPATCH-UNWIRED.md` at line 56, Update the fenced
code block around the ASCII chain diagram in DOC04-WORKROOM-DISPATCH-UNWIRED.md
to specify the text language, using a text fence while preserving the diagram
content.

Source: Linters/SAST tools

[coderabbitai]

📐 Maintainability & Code Quality | 🟡 Minor | ⚡ Quick win

🧩 Analysis chain

🏁 Script executed:

#!/bin/bash
# Does WorkroomDispatchUnavailable still exist? Is the approve route mounted? Where does post_meeting/dispatch.py live?
rg -n 'WorkroomDispatchUnavailable' -g '*.py'
rg -n 'install_approve_route' -g '*.py'
fd -t f 'dispatch.py' --full-path -x echo {}
rg -n 'make_dispatch_workroom_server|dispatch_workroom_tool' -g '*.py'

Repository: dakshparikh12/proxy

Length of output: 157

---

🏁 Script executed:

#!/bin/bash
set -u
printf '%s\n' '--- candidate files ---'
git ls-files | rg '(^|/)(plan_approval_route\.py|dispatch\.py|live_dispatch_deadend\.py|test_seam2_plan_approval\.py|DOC04-WORKROOM-DISPATCH-UNWIRED\.md|dispatch\.py)$'
printf '%s\n' '--- target symbols and route wiring ---'
rg -n -C 3 'WorkroomDispatchUnavailable|install_approve_route|dispatch_workroom_tool|make_dispatch_workroom_server|no dispatcher is configured|Closing it|Closing It' \
  services control_plane harness tests docs 2>/dev/null || true

Repository: dakshparikh12/proxy

Length of output: 20086

---

🏁 Script executed:

#!/bin/bash
set -u
printf '%s\n' '--- candidate files ---'
git ls-files | rg '(^|/)(plan_approval_route\.py|dispatch\.py|live_dispatch_deadend\.py|test_seam2_plan_approval\.py|DOC04-WORKROOM-DISPATCH-UNWIRED\.md)$'
printf '%s\n' '--- target symbols and route wiring ---'
rg -n -C 3 'WorkroomDispatchUnavailable|install_approve_route|dispatch_workroom_tool|make_dispatch_workroom_server|no dispatcher is configured|Closing it|Closing It' \
  services control_plane harness tests docs 2>/dev/null || true

Repository: dakshparikh12/proxy

Length of output: 20086

---

🏁 Script executed:

#!/bin/bash
set -u
printf '%s\n' '--- gap document ---'
sed -n '88,152p' docs/gaps/DOC04-WORKROOM-DISPATCH-UNWIRED.md
printf '%s\n' '--- dispatcher implementation ---'
sed -n '1,28p;120,270p' services/control-plane/src/control_plane/post_meeting/wire.py
printf '%s\n' '--- dispatch wrapper ---'
sed -n '330,452p' services/control-plane/src/control_plane/dispatch.py
printf '%s\n' '--- route and application wiring ---'
sed -n '1,30p;112,145p;155,205p' services/control-plane/src/control_plane/plan_approval_route.py
sed -n '185,215p' services/control-plane/src/control_plane/app.py
printf '%s\n' '--- path/import references ---'
rg -n 'harness\.dispatch|harness\.post_meeting\.dispatch|control_plane\.post_meeting\.dispatch|post_meeting\.wire|make_plan_dispatcher' \
  services tests docs 2>/dev/null || true

Repository: dakshparikh12/proxy

Length of output: 27841

---

Update the DOC04 gap statement and closure steps.

• Line 94: The route no longer raises WorkroomDispatchUnavailable. Without a dispatcher, it records the approval and returns 202 dispatch_blocked with "no dispatcher is configured" in detail.
• Line 95: Replace harness/post_meeting/dispatch.py with services/control-plane/src/control_plane/post_meeting/dispatch.py.
• Steps 1–2: make_dispatch_workroom_server already exists in services/control-plane/src/control_plane/dispatch.py, and make_plan_dispatcher already composes dispatch_workroom, SessionDriver, and run_and_notify in services/control-plane/src/control_plane/post_meeting/wire.py. Describe the remaining work as mounting the existing server, wiring the dispatcher into the application, and adding the required end-to-end proof.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@docs/gaps/DOC04-WORKROOM-DISPATCH-UNWIRED.md` around lines 94 - 96, Update
DOC04-WORKROOM-DISPATCH-UNWIRED.md to state that the approval route records the
approval and returns 202 dispatch_blocked with “no dispatcher is configured”
when no dispatcher exists, rather than raising WorkroomDispatchUnavailable.
Replace the dispatch module reference with
services/control-plane/src/control_plane/post_meeting/dispatch.py, and revise
closure steps 1–2 to reference the existing make_dispatch_workroom_server and
make_plan_dispatcher implementations, focusing remaining work on mounting the
server, wiring the dispatcher into the application, and adding end-to-end proof.

Source: Linters/SAST tools

[coderabbitai]

📐 Maintainability & Code Quality | 🟡 Minor | ⚡ Quick win

Correct the patch count: the pack contains nine patches, not eight.

Line 5 and line 22 state "Eight patches". The body defines P1 through P9. Line 133 repeats "five of the eight patches". Update all three places so the checklist count matches the content.

📝 Proposed fix

-**Read this first.** Four decisions need a founder call (Part 1). Eight patches then apply mechanically (Part 2).
+**Read this first.** Four decisions need a founder call (Part 1). Nine patches then apply mechanically (Part 2).

-## Part 2 · The eight patches
+## Part 2 · The nine patches

Also applies to: 22-22

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@product/v0-spec/AMENDMENTS-06-07.md` at line 5, Update the patch-count
references in the document introduction and checklist summary: change all three
occurrences claiming eight patches—including the line 5 and line 22 statements
and the “five of the eight patches” wording near line 133—to reflect the nine
patches defined by P1 through P9. Do not modify any other content.

[coderabbitai]

📐 Maintainability & Code Quality | 🟡 Minor | ⚡ Quick win

Conform the §2 DDL comment to the P10 ruling.

Line 224 rules that scope_id holds the meeting_id and never the workroom task_id. The §2 DDL comment at line 72 still reads -- meeting_id | task_id. This file wins over every other doc, so the stale comment now contradicts the ruling one section above it. Extend the P10 instruction to also correct that comment.

📝 Proposed fix

-  scope_id          text NOT NULL,          -- meeting_id | task_id
+  scope_id          text NOT NULL,          -- meeting_id ONLY (P10); the task id lives in operation_type

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@product/v0-spec/CANONICAL-DECISIONS.md` at line 224, Update the §2 DDL
comment referenced near the `meeting_id` schema definition to remove the stale
`task_id` association and state that `scope_id` contains the `meeting_id` only.
Keep the existing P10 ruling and UUID/type requirements unchanged, including the
documented `meeting_id::text` cast at the atomic claim call site.

[coderabbitai]

🩺 Stability & Availability | 🔴 Critical | ⚡ Quick win

🧩 Analysis chain

🏁 Script executed:

#!/bin/bash
# Find every caller of dispatch_workroom and check how each detects the gate outcome.
set -uo pipefail

rg -nP --type=py -C 10 '\bdispatch_workroom\s*\(' services tests

# The DispatchDecision shape: does it carry a run_id at all?
rg -nP --type=py -C 12 'class DispatchDecision' libs services

Repository: dakshparikh12/proxy

Length of output: 28412

---

🏁 Script executed:

#!/bin/bash
set -euo pipefail

printf '%s\n' '--- dispatch_workroom contract and implementation ---'
sed -n '130,225p' services/control-plane/src/control_plane/dispatch.py

printf '%s\n' '--- affected control-plane caller ---'
sed -n '340,430p' services/control-plane/src/control_plane/dispatch.py

printf '%s\n' '--- result types and cost gate ---'
sed -n '1,115p' libs/ops/src/ops/cost.py
rg -n -C 12 'estimate_usd|DispatchDecision|dispatched|WorkroomHandle|return decision|return handle' \
  services/control-plane/src/control_plane/dispatch.py libs/ops/src/ops/cost.py

printf '%s\n' '--- tests for gated dispatch ---'
sed -n '235,325p' tests/doc04/test_bundle_dispatch.py

Repository: dakshparikh12/proxy

Length of output: 42993

---

🏁 Script executed:

#!/bin/bash
set -euo pipefail

python3 - <<'PY'
from pathlib import Path
import ast

dispatch_path = Path("services/control-plane/src/control_plane/dispatch.py")
cost_path = Path("libs/ops/src/ops/cost.py")

dispatch = ast.parse(dispatch_path.read_text())
cost = ast.parse(cost_path.read_text())

def find_function(tree, name):
    for node in ast.walk(tree):
        if isinstance(node, (ast.FunctionDef, ast.AsyncFunctionDef)) and node.name == name:
            return node
    raise SystemExit(f"not found by this method: {name}")

fn = find_function(dispatch, "dispatch_workroom")
tool = find_function(dispatch, "dispatch_workroom_tool")
gate = find_function(cost, "dispatch_workroom")

# Confirm the successful gated path returns the decision after claiming a row.
returns = [
    (node.lineno, ast.unparse(node.value))
    for node in ast.walk(fn)
    if isinstance(node, ast.Return)
]
assert any("decision" == value and line >= 199 for line, value in returns), returns
assert any("handle" == value and line >= 196 for line, value in returns), returns

# Confirm the tool reads run_id directly and treats absence as a budget refusal.
tool_source = ast.get_source_segment(dispatch_path.read_text(), tool)
assert tool_source is not None
assert "getattr(outcome, 'run_id', None)" in tool_source
assert "if run_id is None" in tool_source

# Confirm DispatchDecision has no run_id field and the cost gate's success result
# contains dispatched=True but no run_id.
decision = next(
    node for node in ast.walk(cost)
    if isinstance(node, ast.ClassDef) and node.name == "DispatchDecision"
)
fields = {
    node.targets[0].id
    for node in decision.body
    if isinstance(node, ast.AnnAssign)
    and isinstance(node.target, ast.Name)
}
assert "dispatched" in fields
assert "run_id" not in fields

gate_source = ast.get_source_segment(cost_path.read_text(), gate)
assert gate_source is not None
assert "dispatched=True" in gate_source
assert "run_id" not in gate_source

print("resolved: successful gated dispatch returns DispatchDecision(dispatched=True) without run_id")
print("resolved: the tool interprets missing run_id as a budget refusal")
print("resolved: _claim_workroom runs before the successful gated return")
PY

Repository: dakshparikh12/proxy

Length of output: 246

---

🏁 Script executed:

#!/bin/bash
set -euo pipefail

python3 - <<'PY'
from pathlib import Path
import ast

dispatch_file = Path("services/control-plane/src/control_plane/dispatch.py")
cost_file = Path("libs/ops/src/ops/cost.py")
dispatch_text = dispatch_file.read_text()
cost_text = cost_file.read_text()
dispatch_tree = ast.parse(dispatch_text)
cost_tree = ast.parse(cost_text)

def functions(tree, name):
    return [
        node for node in ast.walk(tree)
        if isinstance(node, (ast.FunctionDef, ast.AsyncFunctionDef))
        and node.name == name
    ]

dispatch_fn = functions(dispatch_tree, "dispatch_workroom")[0]
tool_fn = functions(dispatch_tree, "dispatch_workroom_tool")[0]
gate_fn = functions(cost_tree, "dispatch_workroom")[0]

returns = [
    (node.lineno, ast.dump(node.value, include_attributes=False))
    for node in ast.walk(dispatch_fn)
    if isinstance(node, ast.Return)
]
assert any(
    isinstance(node.value, ast.Name) and node.value.id == "decision" and node.lineno >= 199
    for node in ast.walk(dispatch_fn)
    if isinstance(node, ast.Return)
), returns
assert any(
    isinstance(node.value, ast.Name) and node.value.id == "handle" and node.lineno >= 196
    for node in ast.walk(dispatch_fn)
    if isinstance(node, ast.Return)
), returns

# Locate getattr(outcome, "run_id", None) structurally.
run_id_getattr = [
    node for node in ast.walk(tool_fn)
    if isinstance(node, ast.Call)
    and isinstance(node.func, ast.Name)
    and node.func.id == "getattr"
    and len(node.args) >= 2
    and isinstance(node.args[0], ast.Name)
    and node.args[0].id == "outcome"
    and isinstance(node.args[1], ast.Constant)
    and node.args[1].value == "run_id"
]
assert run_id_getattr, "tool does not read outcome.run_id"
assert any(
    isinstance(node, ast.If)
    and isinstance(node.test, ast.Compare)
    and any(
        isinstance(op, ast.Is) for op in node.test.ops
    )
    and isinstance(node.test.comparators[0], ast.Constant)
    and node.test.comparators[0].value is None
    for node in ast.walk(tool_fn)
), "tool has no run_id-is-None branch"

decision_cls = next(
    node for node in ast.walk(cost_tree)
    if isinstance(node, ast.ClassDef) and node.name == "DispatchDecision"
)
fields = {
    node.target.id
    for node in decision_cls.body
    if isinstance(node, ast.AnnAssign)
    and isinstance(node.target, ast.Name)
}
assert "dispatched" in fields
assert "run_id" not in fields

# The gate's successful return sets dispatched=True, while the async dispatcher
# returns decision after the row claim.
assert any(
    isinstance(node, ast.keyword)
    and node.arg == "dispatched"
    and isinstance(node.value, ast.Constant)
    and node.value.value is True
    for node in ast.walk(gate_fn)
), "gate has no dispatched=True result"

claim_calls = [
    node for node in ast.walk(dispatch_fn)
    if isinstance(node, ast.Call)
    and isinstance(node.func, ast.Name)
    and node.func.id == "_claim_workroom_row"
]
decision_returns = [
    node for node in ast.walk(dispatch_fn)
    if isinstance(node, ast.Return)
    and isinstance(node.value, ast.Name)
    and node.value.id == "decision"
]
assert claim_calls and decision_returns
assert claim_calls[0].lineno < decision_returns[-1].lineno

print("resolved: successful gated dispatch returns DispatchDecision(dispatched=True) without run_id")
print("resolved: the tool treats missing outcome.run_id as refusal")
print("resolved: the row claim occurs before the successful gated return")
PY

Repository: dakshparikh12/proxy

Length of output: 246

---

🏁 Script executed:

#!/bin/bash
set -euo pipefail

python3 - <<'PY'
from pathlib import Path
import ast

dispatch_file = Path("services/control-plane/src/control_plane/dispatch.py")
cost_file = Path("libs/ops/src/ops/cost.py")
dispatch_tree = ast.parse(dispatch_file.read_text())
cost_tree = ast.parse(cost_file.read_text())

def fn(tree, name):
    return next(
        node for node in ast.walk(tree)
        if isinstance(node, (ast.FunctionDef, ast.AsyncFunctionDef))
        and node.name == name
    )

dispatch_fn = fn(dispatch_tree, "dispatch_workroom")
tool_fn = fn(dispatch_tree, "dispatch_workroom_tool")
gate_fn = fn(cost_tree, "dispatch_workroom")

claim_lines = sorted(
    node.lineno for node in ast.walk(dispatch_fn)
    if isinstance(node, ast.Call)
    and isinstance(node.func, ast.Name)
    and node.func.id == "_claim_workroom_row"
)
decision_return_lines = sorted(
    node.lineno for node in ast.walk(dispatch_fn)
    if isinstance(node, ast.Return)
    and isinstance(node.value, ast.Name)
    and node.value.id == "decision"
)
handle_return_lines = sorted(
    node.lineno for node in ast.walk(dispatch_fn)
    if isinstance(node, ast.Return)
    and isinstance(node.value, ast.Name)
    and node.value.id == "handle"
)

decision_cls = next(
    node for node in ast.walk(cost_tree)
    if isinstance(node, ast.ClassDef) and node.name == "DispatchDecision"
)
fields = sorted(
    node.target.id for node in decision_cls.body
    if isinstance(node, ast.AnnAssign)
    and isinstance(node.target, ast.Name)
)

run_id_getattr_lines = sorted(
    node.lineno for node in ast.walk(tool_fn)
    if isinstance(node, ast.Call)
    and isinstance(node.func, ast.Name)
    and node.func.id == "getattr"
    and len(node.args) >= 2
    and isinstance(node.args[0], ast.Name)
    and node.args[0].id == "outcome"
    and isinstance(node.args[1], ast.Constant)
    and node.args[1].value == "run_id"
)

dispatch_true_lines = sorted(
    node.lineno for node in ast.walk(gate_fn)
    if isinstance(node, ast.keyword)
    and node.arg == "dispatched"
    and isinstance(node.value, ast.Constant)
    and node.value.value is True
)

print("claim_lines:", claim_lines)
print("decision_return_lines:", decision_return_lines)
print("handle_return_lines:", handle_return_lines)
print("DispatchDecision fields:", fields)
print("tool outcome.run_id getattr lines:", run_id_getattr_lines)
print("gate dispatched=True keyword lines:", dispatch_true_lines)

print("resolved: DispatchDecision has no run_id:", "run_id" not in fields)
print("resolved: tool reads outcome.run_id:", bool(run_id_getattr_lines))
print("resolved: gate can return dispatched=True:", bool(dispatch_true_lines))
print(
    "resolved: successful gated return follows claim:",
    bool(claim_lines and decision_return_lines and min(claim_lines) < max(decision_return_lines)),
)
PY

Repository: dakshparikh12/proxy

Length of output: 594

---

Return the WorkroomHandle after a successful cost gate.

A successful gated call claims the operation_runs row, then returns DispatchDecision(dispatched=True) without run_id. The tool treats the missing run_id as a budget refusal, skips run_task, and leaves the row running.

Return the handle on the successful path, or resolve it before the tool checks run_id.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@services/control-plane/src/control_plane/dispatch.py` around lines 391 - 411,
The successful cost-gated path in the dispatch flow must expose the claimed run
identifier. Update dispatch_workroom and the surrounding handling so a
successful call returns or resolves a WorkroomHandle containing run_id before
the existing run_id check; preserve the None path for budget refusals, then
continue to run_task and run_and_notify with the resolved identifier.

[coderabbitai]

🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

One banned token is a regex used in a substring test, so it never matches.

"class .*queue" is a regular expression. Line 129 performs a substring test against lowered source, so this element can never match and asserts nothing. Use re.search for the pattern, or replace it with a literal.

🐛 Proposed fix

+    import re
+
     lowered = src.lower()
-    for token in ("class .*queue", "sandboxprovider(", "e2b.", "scheduler("):
+    assert not re.search(r"class \w*queue", lowered), "a local queue class was defined"
+    for token in ("sandboxprovider(", "e2b.", "scheduler("):
         assert token not in lowered

📝 Committable suggestion

‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.

Suggested change

	lowered = src.lower()
	for token in ("class .*queue", "sandboxprovider(", "e2b.", "scheduler("):
	assert token not in lowered
	import re
	lowered = src.lower()
	assert not re.search(r"class \w*queue", lowered), "a local queue class was defined"
	for token in ("sandboxprovider(", "e2b.", "scheduler("):
	assert token not in lowered

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tests/doc07/test_b6_dispatch.py` around lines 127 - 129, Update the
banned-token check in the test around lowered source so the “class .*queue”
pattern is evaluated as a regular expression via re.search, or replace it with
the intended literal token. Preserve the existing checks for “sandboxprovider(”,
“e2b.”, and “scheduler(”.

[coderabbitai]

📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Both tests read source files through a hard-coded repo-root-relative path. The assertions pass only when pytest runs from the repository root, and they break silently at the next directory move. Resolve the path from the imported module instead, so the check follows the module wherever it lives.

• tests/doc07/test_b7_report.py#L233-L235: import control_plane.post_meeting.report and read pathlib.Path(report.__file__) instead of the literal services/control-plane/src/... path.
• tests/doc07/test_seam2_plan_approval.py#L227-L229: import control_plane.plan_approval_route and read pathlib.Path(m.__file__); the module is already imported at Line 77 in a sibling test.

📍 Affects 2 files

• tests/doc07/test_b7_report.py#L233-L235 (this comment)
• tests/doc07/test_seam2_plan_approval.py#L227-L229

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tests/doc07/test_b7_report.py` around lines 233 - 235, Replace the hard-coded
source path in tests/doc07/test_b7_report.py:233-235 by importing
control_plane.post_meeting.report and resolving pathlib.Path(report.__file__).
Apply the same change in tests/doc07/test_seam2_plan_approval.py:227-229, using
the existing control_plane.plan_approval_route module import (m) and
pathlib.Path(m.__file__).

[coderabbitai]

🎯 Functional Correctness | 🔵 Trivial | ⚡ Quick win

Compare the sweep row count exactly instead of using endswith.

conn.execute returns a status string such as "UPDATE 1". swept.endswith("1") also matches "UPDATE 11" and "UPDATE 21", and endswith("0") also matches "UPDATE 10". A sweep that touched more rows than intended still passes.

♻️ Proposed change

-    assert swept.endswith("1"), f"the stale run was not swept: {swept!r}"
+    assert swept == "UPDATE 1", f"the stale run was not swept exactly once: {swept!r}"

Apply the same exact comparison at Line 195 ("UPDATE 0") and Line 223 (first == "UPDATE 1", second == "UPDATE 0").

Also applies to: 195-196, 223-223

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tests/doc07/test_integration_claim_recycle.py` at line 157, Replace
suffix-based sweep count assertions around the integration test’s sweep results
with exact status-string comparisons: compare swept to "UPDATE 1", the zero-row
result at the later assertion to "UPDATE 0", and first/second to "UPDATE
1"/"UPDATE 0" respectively. Preserve the existing assertion messages and test
flow.

[coderabbitai]

📐 Maintainability & Code Quality | 🟡 Minor | ⚡ Quick win

🧩 Analysis chain

🏁 Script executed:

#!/bin/bash
# Which migration declares each object, and what are the constraint names?
rg -n 'staged_drafts_status_enum|status_check|planned_at|approved_needs_approver' migrations/versions

Repository: dakshparikh12/proxy

Length of output: 2597

---

🏁 Script executed:

#!/bin/bash
set -eu

printf '%s\n' '--- test module imports and assertions ---'
sed -n '1,145p' tests/doc07/test_integration_db.py
printf '%s\n' '--- migration-related docstring sections ---'
sed -n '195,220p' tests/doc07/test_integration_db.py
sed -n '312,338p' tests/doc07/test_integration_db.py
printf '%s\n' '--- exception references in the module ---'
rg -n 'asyncpg|psycopg|IntegrityError|ForeignKeyViolation|CheckViolation|UniqueViolation|constraint' tests/doc07/test_integration_db.py

Repository: dakshparikh12/proxy

Length of output: 10081

---

🏁 Script executed:

#!/bin/bash
set -eu

printf '%s\n' '--- exact stale docstrings ---'
sed -n '320,332p' tests/doc07/test_integration_db.py
printf '%s\n' '--- migration revision chain and declarations ---'
sed -n '1,75p' migrations/versions/0011_clarify_items.py
sed -n '1,75p' migrations/versions/0012_staged_drafts_status_check.py
sed -n '1,65p' migrations/versions/0013_post_meeting_planned_at.py

Repository: dakshparikh12/proxy

Length of output: 9394

---

Update the stale test documentation.

• Replace psycopg with asyncpg exception terminology. The module uses asyncpg.exceptions.*, including non-integrity RaiseError.
• Change migration 0011 to 0012_staged_drafts_status_check.py.
• Change migration 0012 to 0013_post_meeting_planned_at.py.
• The constraint name remains staged_drafts_status_enum.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tests/doc07/test_integration_db.py` around lines 14 - 17, Update the test
documentation to use asyncpg exception terminology, including references to
asyncpg.exceptions.* and non-integrity RaiseError. Revise the migration
references from 0011 to 0012_staged_drafts_status_check.py and from 0012 to
0013_post_meeting_planned_at.py, while preserving the staged_drafts_status_enum
constraint name.

[coderabbitai]

📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

Replace __import__ with a normal import.

test_b4_plan.py already imports PostMeetingConfig at module level. Use the same import here so the config type is visible to readers and to mypy.

♻️ Proposed refactor

Add the import at the top of the file:

+from control_plane.post_meeting.config import PostMeetingConfig
 from control_plane.post_meeting.plan import expire_stale_plans

Then simplify the call:

     res = await expire_stale_plans(
         rows, store=store, now=second + timedelta(hours=1),
-        config=__import__(
-            "control_plane.post_meeting.config", fromlist=["PostMeetingConfig"]
-        ).PostMeetingConfig(plan_expiry_hours=48),
+        config=PostMeetingConfig(plan_expiry_hours=48),
     )

📝 Committable suggestion

‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.

Suggested change

	res = await expire_stale_plans(
	rows, store=store, now=second + timedelta(hours=1),
	config=__import__(
	"control_plane.post_meeting.config", fromlist=["PostMeetingConfig"]
	).PostMeetingConfig(plan_expiry_hours=48),
	)
	assert res.expired == []
	from control_plane.post_meeting.config import PostMeetingConfig
	from control_plane.post_meeting.plan import expire_stale_plans
	res = await expire_stale_plans(
	rows, store=store, now=second + timedelta(hours=1),
	config=PostMeetingConfig(plan_expiry_hours=48),
	)
	assert res.expired == []

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tests/doc07/test_owner_actions.py` around lines 151 - 157, Replace the
dynamic __import__ expression in the expire_stale_plans call with a module-level
import of PostMeetingConfig, matching the existing pattern in test_b4_plan.py.
Use the imported PostMeetingConfig directly when constructing the config
argument, leaving the test behavior unchanged.