fix(lifecycle): let a PM recover its rejected coordination task from needs_revision

The in-path PR-review gate created a deadlock: when an assembled cell→root /
root→master PR fails the gate (pr_fail) — or qa_fail / ceo_reject fires — the
PM-owned coordination task lands in needs_revision, which was developer-claim-
only. So the task had no actor and no exit but cancel, and the cell PM escalated
in a loop (8KB of [ESCALATED] dev_notes on one task). Pre-gate, the PM simply
re-delegated from in_progress; the gate routed the failure through the PM's own
task instead.

Add NEEDS_REVISION to the CELL_PM / MAIN_PM claim rules so the PM re-claims via
i_will_plan, revises the plan, and re-delegates the fixes — pr_fail/qa_fail
already reassign the failed task to its owning PM and the revision dispatcher
re-spawns it; the claim rule was the only missing piece.

Scope is by give_me_work routing (offers only the caller's own assigned tasks),
the same mechanism that scopes a developer's leaf-revision — NOT a gateway-only
ownership gate, which would violate the spec=gateway parity invariant and can't
use task_type anyway (main-PM coordination roots can be code-typed). Regenerates
panel/lib/lifecycle.json.
This commit is contained in:
Renn F
2026-06-21 06:09:59 +02:00
parent 3ff6967b9d
commit 8463808616
5 changed files with 127 additions and 5 deletions
+2
View File
@@ -2,6 +2,7 @@
"claim_rules": {
"auditor": [],
"cell_pm": [
"needs_revision",
"pending"
],
"ceo": [],
@@ -15,6 +16,7 @@
],
"head_marketing": [],
"main_pm": [
"needs_revision",
"pending"
],
"pr_reviewer": [
+14 -2
View File
@@ -666,8 +666,20 @@ CLAIM_RULES: dict[Role, frozenset[Status]] = {
Role.DEVELOPER: frozenset({Status.PENDING, Status.NEEDS_REVISION}),
Role.QA: frozenset({Status.AWAITING_QA}),
Role.DOCUMENTER: frozenset({Status.PENDING, Status.AWAITING_DOCUMENTATION}),
Role.CELL_PM: frozenset({Status.PENDING}),
Role.MAIN_PM: frozenset({Status.PENDING}),
# PMs re-claim NEEDS_REVISION to recover a rejected coordination task: when
# an assembled cell→root / root→master PR fails the in-path gate (pr_fail),
# QA fails a planning task, or the CEO rejects (ceo_reject), the task lands
# in NEEDS_REVISION. Before this, that state was developer-claim-only, so a
# PM-owned coordination task had NO actor and no exit but cancel — the cell
# PM escalated in a loop (the in-path PR-review gate introduced this path;
# pre-gate, the PM re-delegated from in_progress). The PM now re-claims via
# i_will_plan, revises the plan, and re-delegates the fixes. pr_fail/qa_fail
# reassign the task to its owning PM, so this is scoped by the SAME mechanism
# that scopes a developer's leaf-revision: give_me_work only ever offers an
# agent its own assigned tasks. (A per-instance ownership gate at the gateway
# would diverge from this spec — the parity invariant forbids that.)
Role.CELL_PM: frozenset({Status.PENDING, Status.NEEDS_REVISION}),
Role.MAIN_PM: frozenset({Status.PENDING, Status.NEEDS_REVISION}),
Role.PRODUCT_OWNER: frozenset(),
Role.HEAD_MARKETING: frozenset(),
Role.AUDITOR: frozenset(),
@@ -1067,6 +1067,11 @@ class Choreographer:
Per-role claim authority (CLAIM_RULES) is enforced inside
spec.can_invoke_action when action == "claim", called by
can_invoke_intent, so no separate spec.can_claim call is needed.
A PM recovering a NEEDS_REVISION coordination task is scoped by
give_me_work routing (it only ever offers an agent its OWN assigned
tasks), exactly as developer leaf-revisions are not by a gateway-only
ownership gate, which would diverge from the spec (the parity invariant
requires gateway authorization == spec authorization).
"""
t, briefing, role_str = ctx.task, ctx.briefing, ctx.role_str
verb_name = ctx.verb_name
+10 -3
View File
@@ -406,7 +406,10 @@ def test_qa_pass_self_review_blocks() -> None:
def test_claim_rules_match_pre_gateway_table() -> None:
"""PERMISSIONS.md "What Each Role Can Claim From" — exact match.
PMs claim from PENDING only; BACKLOG PENDING is a separate `activate`
PMs claim from PENDING and NEEDS_REVISION the latter to recover a rejected
coordination task (pr_fail / qa_fail / ceo_reject) by re-planning and
re-delegating fixes (scoped by give_me_work routing, which offers only the
caller's own assigned tasks). BACKLOG → PENDING is a separate `activate`
action (strict transitions; no implicit activate-on-claim).
"""
assert spec.CLAIM_RULES[spec.Role.DEVELOPER] == frozenset(
@@ -416,8 +419,12 @@ def test_claim_rules_match_pre_gateway_table() -> None:
assert spec.CLAIM_RULES[spec.Role.DOCUMENTER] == frozenset(
{spec.Status.PENDING, spec.Status.AWAITING_DOCUMENTATION}
)
assert spec.CLAIM_RULES[spec.Role.CELL_PM] == frozenset({spec.Status.PENDING})
assert spec.CLAIM_RULES[spec.Role.MAIN_PM] == frozenset({spec.Status.PENDING})
assert spec.CLAIM_RULES[spec.Role.CELL_PM] == frozenset(
{spec.Status.PENDING, spec.Status.NEEDS_REVISION}
)
assert spec.CLAIM_RULES[spec.Role.MAIN_PM] == frozenset(
{spec.Status.PENDING, spec.Status.NEEDS_REVISION}
)
def test_team_rules_pin_team_for_seeded_agents() -> None:
@@ -0,0 +1,96 @@
"""PM recovery of a rejected coordination task from needs_revision.
The in-path PR-review gate (and qa_fail / ceo_reject) can land a PM-owned
coordination/assembled task in ``needs_revision``. That state used to be
developer-claim-only, so the task had no actor and no exit but ``cancel`` the
cell PM escalated in a loop. Now a PM re-claims its rejected task
(``i_will_plan``), revises the plan, and re-delegates the fixes.
Scope: ``pr_fail`` / ``qa_fail`` reassign the task to its owning PM, and
``give_me_work`` only ever offers an agent its OWN assigned tasks so a PM is
never handed a developer's leaf, exactly as a developer is never handed a PM's
coordination root. (The scope lives in routing, not a gateway-only ownership
gate, which would break the spec=gateway parity invariant.)
"""
from __future__ import annotations
from types import SimpleNamespace
from unittest.mock import AsyncMock, MagicMock
from uuid import uuid4
import pytest
from roboco.foundation.policy import lifecycle as spec
from roboco.services.gateway.choreographer import Choreographer, ChoreographerDeps
# --------------------------------------------------------------------------- #
# Lifecycle authority: PMs may now claim needs_revision
# --------------------------------------------------------------------------- #
def test_pm_can_claim_needs_revision() -> None:
task = SimpleNamespace(
status="needs_revision", task_type="planning", assigned_to=None
)
for role in (spec.Role.CELL_PM, spec.Role.MAIN_PM):
assert spec.can_invoke_action(role, "claim", task).allowed
# Developers still own leaf revisions; QA / documenter still cannot claim it.
assert spec.can_invoke_action(spec.Role.DEVELOPER, "claim", task).allowed
assert not spec.can_invoke_action(spec.Role.QA, "claim", task).allowed
assert not spec.can_invoke_action(spec.Role.DOCUMENTER, "claim", task).allowed
def test_pm_claim_needs_revision_works_for_code_typed_root() -> None:
# Main-PM coordination roots can be code-typed, so the claim must NOT be
# task_type-gated — authority is status-based, scoped by routing.
task = SimpleNamespace(status="needs_revision", task_type="code", assigned_to=None)
assert spec.can_invoke_action(spec.Role.MAIN_PM, "claim", task).allowed
# --------------------------------------------------------------------------- #
# Routing scope: give_me_work offers a PM its OWN rejected coordination task
# --------------------------------------------------------------------------- #
def _make_deps(task_svc: AsyncMock) -> ChoreographerDeps:
repo = AsyncMock()
for m in (
"list_unread_a2a",
"list_unread_mentions",
"list_pending_notifications",
"task_metadata_gaps",
"recent_team_activity",
"blockers_in_lane",
"journal_highlights_for_task",
"company_goals",
):
getattr(repo, m).return_value = []
return ChoreographerDeps(
task=task_svc,
work_session=AsyncMock(),
git=AsyncMock(),
a2a=AsyncMock(),
journal=AsyncMock(),
audit=AsyncMock(),
evidence_repo=repo,
)
@pytest.mark.asyncio
async def test_give_me_work_offers_pm_its_needs_revision_task() -> None:
pm_id, task_id = uuid4(), uuid4()
task = MagicMock(
id=task_id, status="needs_revision", team="backend", dependency_ids=[]
)
task_svc = AsyncMock()
task_svc.list_pending_for_agent.return_value = []
task_svc.list_assigned_for_agent.return_value = [task]
task_svc.agent_for.return_value = MagicMock(role="cell_pm", team="backend")
c = Choreographer(_make_deps(task_svc))
env = await c.give_me_work(pm_id)
body = env.as_dict()
assert body["task_id"] == str(task_id)
# The PM is told to re-plan (revise) the rejected task, not a dev verb.
assert "i_will_plan" in body["next"]