@linimin/pi-letscook 0.1.29 → 0.1.31

package/scripts/context-proposal-test.sh

@@ -120,30 +120,57 @@ PY
 # No workflow yet: /cook with no goal should infer from recent discussion through analyst output.
 SESSION_ONE="$TMPDIR/session-one.jsonl"
 DISCUSSION_ONE="$DISCUSSION_ZERO"
-ANALYST_OUTPUT_ONE='{"mission":"Remove the completion status line while keeping the completion widget.","scope":["Keep the non-running completion widget.","Suppress the widget while a completion role is active."],"constraints":["Do not reintroduce any other completion status surface."],"acceptance":["Update README to match the shipped behavior.","Keep observability regression coverage truthful."],"confidence":0.94}'
+ANALYST_OUTPUT_ONE='{"mission":"Remove the completion status line while keeping the completion widget.","scope":["Keep the non-running completion widget.","Suppress the widget while a completion role is active."],"constraints":["Do not reintroduce any other completion status surface."],"acceptance":["Update README to match the shipped behavior.","Keep observability regression coverage truthful."],"critique":["Keep critique separate from the mission anchor so startup analysis does not rewrite the workflow goal."],"risks":["Stale widget-removal discussion could broaden the startup plan if it gets treated as mission text."],"task_type":"completion-workflow","evaluation_profile":"completion-rubric-v1","possible_noise":["older widget restyle ideas"],"confidence":0.94}'
+DISCUSSION_SNAPSHOT_ONE="$TMPDIR/context-proposal-discussion-hints.json"
 write_session "$SESSION_ONE" "$ROOT" "$DISCUSSION_ONE"
 
 PI_COMPLETION_CONTEXT_PROPOSAL_ACTION=accept \
 PI_COMPLETION_CONTEXT_PROPOSAL_ANALYST_OUTPUT="$ANALYST_OUTPUT_ONE" \
+PI_COMPLETION_TEST_CONTEXT_PROPOSAL_PATH="$DISCUSSION_SNAPSHOT_ONE" \
 PI_COMPLETION_SKIP_DRIVER_KICKOFF=1 \
 pi --session "$SESSION_ONE" -e "$PKG_ROOT" -p "/cook" >/tmp/pi-completion-context-proposal-bootstrap.out 2>/tmp/pi-completion-context-proposal-bootstrap.err
 
-python3 - <<'PY'
+python3 - "$DISCUSSION_SNAPSHOT_ONE" <<'PY'
 import json
+import sys
 from pathlib import Path
 
 mission = 'Remove the completion status line while keeping the completion widget.'
+expected_task_type = 'completion-workflow'
+expected_eval_profile = 'completion-rubric-v1'
 mission_text = Path('.agent/mission.md').read_text()
+profile = json.loads(Path('.agent/profile.json').read_text())
 state = json.loads(Path('.agent/state.json').read_text())
 plan = json.loads(Path('.agent/plan.json').read_text())
 active = json.loads(Path('.agent/active-slice.json').read_text())
+proposal = json.loads(Path(sys.argv[1]).read_text())
 
 assert mission in mission_text, '.agent/mission.md did not record the analyst-derived mission anchor'
+assert profile['task_type'] == expected_task_type, 'profile.json task_type mismatch after analyst-derived bootstrap'
+assert profile['evaluation_profile'] == expected_eval_profile, 'profile.json evaluation_profile mismatch after analyst-derived bootstrap'
 assert state['mission_anchor'] == mission, 'state.json mission_anchor mismatch after analyst-derived bootstrap'
+assert state['task_type'] == expected_task_type, 'state.json task_type mismatch after analyst-derived bootstrap'
+assert state['evaluation_profile'] == expected_eval_profile, 'state.json evaluation_profile mismatch after analyst-derived bootstrap'
 assert plan['mission_anchor'] == mission, 'plan.json mission_anchor mismatch after analyst-derived bootstrap'
+assert plan['task_type'] == expected_task_type, 'plan.json task_type mismatch after analyst-derived bootstrap'
+assert plan['evaluation_profile'] == expected_eval_profile, 'plan.json evaluation_profile mismatch after analyst-derived bootstrap'
 assert active['mission_anchor'] == mission, 'active-slice.json mission_anchor mismatch after analyst-derived bootstrap'
+assert active['task_type'] == expected_task_type, 'active-slice.json task_type mismatch after analyst-derived bootstrap'
+assert active['evaluation_profile'] == expected_eval_profile, 'active-slice.json evaluation_profile mismatch after analyst-derived bootstrap'
+assert proposal['mission'] == mission, 'discussion-only proposal snapshot should keep the inferred mission anchor'
+assert proposal['analysis']['taskType'] == expected_task_type, 'discussion-only proposal snapshot should expose task_type hints separately'
+assert proposal['analysis']['evaluationProfile'] == expected_eval_profile, 'discussion-only proposal snapshot should expose evaluation_profile hints separately'
+assert proposal['analysis']['critique'] == ['Keep critique separate from the mission anchor so startup analysis does not rewrite the workflow goal.'], 'discussion-only proposal snapshot should preserve critique hints'
+assert proposal['analysis']['risks'] == ['Stale widget-removal discussion could broaden the startup plan if it gets treated as mission text.'], 'discussion-only proposal snapshot should preserve risk hints'
+assert proposal['analysis']['possibleNoise'] == ['older widget restyle ideas'], 'discussion-only proposal snapshot should preserve possible_noise hints'
+assert 'Critique:' not in proposal['goalText'], 'goalText should keep critique separate from mission/scope/constraints/acceptance'
+assert 'Task type:' not in proposal['goalText'], 'goalText should keep task_type hints separate from the mission body'
 assert state['current_phase'] == 'reground', 'state.json current_phase should start at reground after analyst-derived bootstrap'
 assert state['next_mandatory_role'] == 'completion-regrounder', 'next_mandatory_role should start at completion-regrounder after analyst-derived bootstrap'
+assert state['continuation_reason'].startswith('User started workflow via /cook:'), 'initial startup should record the accepted startup routing in continuation_reason'
+assert 'task_type=completion-workflow' in state['continuation_reason'], 'initial startup should persist the selected task_type in continuation_reason'
+assert 'evaluation_profile=completion-rubric-v1' in state['continuation_reason'], 'initial startup should persist the selected evaluation_profile in continuation_reason'
+assert 'Keep critique separate from the mission anchor so startup analysis does not rewrite the workflow goal.' in state['continuation_reason'], 'initial startup should persist the accepted critique outcome in continuation_reason'
 PY
 
 # Completed workflow: /cook with no goal should infer the next round from recent discussion through analyst output.
@@ -164,21 +191,35 @@ import json
 from pathlib import Path
 
 mission = 'Ship the next workflow round for richer context-derived /cook startup.'
+expected_task_type = 'completion-workflow'
+expected_eval_profile = 'completion-rubric-v1'
 mission_text = Path('.agent/mission.md').read_text()
+profile = json.loads(Path('.agent/profile.json').read_text())
 state = json.loads(Path('.agent/state.json').read_text())
 plan = json.loads(Path('.agent/plan.json').read_text())
 active = json.loads(Path('.agent/active-slice.json').read_text())
 
 assert mission in mission_text, '.agent/mission.md did not update to the next-round context-derived mission anchor'
+assert profile['task_type'] == expected_task_type, 'profile.json task_type mismatch after next-round startup'
+assert profile['evaluation_profile'] == expected_eval_profile, 'profile.json evaluation_profile mismatch after next-round startup'
 assert state['mission_anchor'] == mission, 'state.json mission_anchor mismatch after starting the next workflow round'
+assert state['task_type'] == expected_task_type, 'state.json task_type mismatch after starting the next workflow round'
+assert state['evaluation_profile'] == expected_eval_profile, 'state.json evaluation_profile mismatch after starting the next workflow round'
 assert plan['mission_anchor'] == mission, 'plan.json mission_anchor mismatch after starting the next workflow round'
+assert plan['task_type'] == expected_task_type, 'plan.json task_type mismatch after starting the next workflow round'
+assert plan['evaluation_profile'] == expected_eval_profile, 'plan.json evaluation_profile mismatch after starting the next workflow round'
 assert active['mission_anchor'] == mission, 'active-slice.json mission_anchor mismatch after starting the next workflow round'
+assert active['task_type'] == expected_task_type, 'active-slice.json task_type mismatch after starting the next workflow round'
+assert active['evaluation_profile'] == expected_eval_profile, 'active-slice.json evaluation_profile mismatch after starting the next workflow round'
 assert state['current_phase'] == 'reground', 'state.json current_phase should reset to reground for the next workflow round'
 assert state['continuation_policy'] == 'continue', 'continuation_policy should reset to continue for the next workflow round'
 assert state['requires_reground'] is True, 'requires_reground should reset to true for the next workflow round'
 assert state['project_done'] is False, 'project_done should reset to false for the next workflow round'
 assert state['next_mandatory_role'] == 'completion-regrounder', 'next_mandatory_role should reset to completion-regrounder for the next workflow round'
 assert state['continuation_reason'].startswith('User refocused workflow via /cook:'), 'continuation_reason should record the next-round refocus'
+assert 'task_type=completion-workflow' in state['continuation_reason'], 'next-round refocus should persist the selected task_type'
+assert 'evaluation_profile=completion-rubric-v1' in state['continuation_reason'], 'next-round refocus should persist the selected evaluation_profile'
+assert 'critique outcome=accepted critique=none' in state['continuation_reason'], 'next-round refocus should persist that no critique notes were accepted'
 assert plan['plan_basis'] == 'user_refocus', 'plan_basis should reset to user_refocus for the next workflow round'
 assert active['status'] == 'idle', 'active-slice should reset to idle for the next workflow round'
 PY
@@ -200,19 +241,33 @@ import json
 from pathlib import Path
 
 mission = 'Explicit replacement mission for the active workflow.'
+expected_task_type = 'completion-workflow'
+expected_eval_profile = 'completion-rubric-v1'
 mission_text = Path('.agent/mission.md').read_text()
+profile = json.loads(Path('.agent/profile.json').read_text())
 state = json.loads(Path('.agent/state.json').read_text())
 plan = json.loads(Path('.agent/plan.json').read_text())
 active = json.loads(Path('.agent/active-slice.json').read_text())
 
 assert mission in mission_text, '.agent/mission.md did not update to the explicit replacement mission anchor'
+assert profile['task_type'] == expected_task_type, 'profile.json task_type mismatch after explicit-goal replacement'
+assert profile['evaluation_profile'] == expected_eval_profile, 'profile.json evaluation_profile mismatch after explicit-goal replacement'
 assert state['mission_anchor'] == mission, 'state.json mission_anchor mismatch after explicit-goal replacement'
+assert state['task_type'] == expected_task_type, 'state.json task_type mismatch after explicit-goal replacement'
+assert state['evaluation_profile'] == expected_eval_profile, 'state.json evaluation_profile mismatch after explicit-goal replacement'
 assert plan['mission_anchor'] == mission, 'plan.json mission_anchor mismatch after explicit-goal replacement'
+assert plan['task_type'] == expected_task_type, 'plan.json task_type mismatch after explicit-goal replacement'
+assert plan['evaluation_profile'] == expected_eval_profile, 'plan.json evaluation_profile mismatch after explicit-goal replacement'
 assert active['mission_anchor'] == mission, 'active-slice.json mission_anchor mismatch after explicit-goal replacement'
+assert active['task_type'] == expected_task_type, 'active-slice.json task_type mismatch after explicit-goal replacement'
+assert active['evaluation_profile'] == expected_eval_profile, 'active-slice.json evaluation_profile mismatch after explicit-goal replacement'
 assert state['current_phase'] == 'reground', 'current_phase should reset to reground after explicit-goal replacement'
 assert state['continuation_policy'] == 'continue', 'continuation_policy should stay continue after explicit-goal replacement'
 assert state['next_mandatory_role'] == 'completion-regrounder', 'next role should reset to completion-regrounder after explicit-goal replacement'
 assert state['continuation_reason'].startswith('User refocused workflow via /cook:'), 'continuation_reason should record the explicit-goal replacement'
+assert 'task_type=completion-workflow' in state['continuation_reason'], 'explicit-goal replacement should persist the selected task_type'
+assert 'evaluation_profile=completion-rubric-v1' in state['continuation_reason'], 'explicit-goal replacement should persist the selected evaluation_profile'
+assert 'critique outcome=accepted critique=none' in state['continuation_reason'], 'explicit-goal replacement should persist the accepted critique outcome even when no critique was derived'
 assert 'Preserve the richer proposal structure from discussion.' not in state['continuation_reason'], 'session scope should not be merged when analyst output is unavailable'
 assert 'Keep explicit goals as the mission anchor when they conflict with earlier text.' not in state['continuation_reason'], 'session constraints should not be merged when analyst output is unavailable'
 assert 'Refresh canonical state from the replacement mission.' not in state['continuation_reason'], 'session acceptance should not be merged when analyst output is unavailable'
@@ -226,35 +281,60 @@ mark_done
 
 SESSION_FOUR="$TMPDIR/session-four.jsonl"
 DISCUSSION_FOUR=$'Scope:\n- Add session-only scope.\n- Restyle widget.\nConstraints:\n- Keep rules.\nAcceptance:\n- Add test.'
-EXPLICIT_GOAL_FOUR=$'Mission: Filter scope by mission.\nScope:\n- Keep explicit scope.'
+EXPLICIT_GOAL_FOUR=$'Mission: Filter scope by mission.\nScope:\n- Keep explicit scope.\nCritique:\n- Keep critique notes separate from the mission anchor.\nRisks:\n- Session-only scope could leak into the next workflow round.\nTask type: completion-workflow\nEvaluation profile: completion-rubric-v1'
+EXPLICIT_SNAPSHOT_FOUR="$TMPDIR/context-proposal-explicit-hints.json"
 write_session "$SESSION_FOUR" "$ROOT" "$DISCUSSION_FOUR"
 
 PI_COMPLETION_CONTEXT_PROPOSAL_ACTION=accept \
 PI_COMPLETION_DISABLE_CONTEXT_PROPOSAL_ANALYST=1 \
+PI_COMPLETION_TEST_CONTEXT_PROPOSAL_PATH="$EXPLICIT_SNAPSHOT_FOUR" \
 PI_COMPLETION_SKIP_DRIVER_KICKOFF=1 \
 pi --session "$SESSION_FOUR" -e "$PKG_ROOT" -p "/cook $EXPLICIT_GOAL_FOUR" >/tmp/pi-completion-context-proposal-done-goal.out 2>/tmp/pi-completion-context-proposal-done-goal.err
 
-python3 - <<'PY'
+python3 - "$EXPLICIT_SNAPSHOT_FOUR" <<'PY'
 import json
+import sys
 from pathlib import Path
 
 mission = 'Filter scope by mission.'
+expected_task_type = 'completion-workflow'
+expected_eval_profile = 'completion-rubric-v1'
 mission_text = Path('.agent/mission.md').read_text()
+profile = json.loads(Path('.agent/profile.json').read_text())
 state = json.loads(Path('.agent/state.json').read_text())
 plan = json.loads(Path('.agent/plan.json').read_text())
 active = json.loads(Path('.agent/active-slice.json').read_text())
+proposal = json.loads(Path(sys.argv[1]).read_text())
 continuation_reason = state['continuation_reason']
 
 assert mission in mission_text, '.agent/mission.md did not update to the explicit next-round mission anchor'
+assert profile['task_type'] == expected_task_type, 'profile.json task_type mismatch after explicit-goal next-round start'
+assert profile['evaluation_profile'] == expected_eval_profile, 'profile.json evaluation_profile mismatch after explicit-goal next-round start'
 assert state['mission_anchor'] == mission, 'state.json mission_anchor mismatch after explicit-goal next-round start'
+assert state['task_type'] == expected_task_type, 'state.json task_type mismatch after explicit-goal next-round start'
+assert state['evaluation_profile'] == expected_eval_profile, 'state.json evaluation_profile mismatch after explicit-goal next-round start'
 assert plan['mission_anchor'] == mission, 'plan.json mission_anchor mismatch after explicit-goal next-round start'
+assert plan['task_type'] == expected_task_type, 'plan.json task_type mismatch after explicit-goal next-round start'
+assert plan['evaluation_profile'] == expected_eval_profile, 'plan.json evaluation_profile mismatch after explicit-goal next-round start'
 assert active['mission_anchor'] == mission, 'active-slice.json mission_anchor mismatch after explicit-goal next-round start'
+assert active['task_type'] == expected_task_type, 'active-slice.json task_type mismatch after explicit-goal next-round start'
+assert active['evaluation_profile'] == expected_eval_profile, 'active-slice.json evaluation_profile mismatch after explicit-goal next-round start'
+assert proposal['mission'] == mission, 'explicit-goal proposal snapshot should preserve the explicit mission anchor'
+assert proposal['analysis']['taskType'] == expected_task_type, 'explicit-goal proposal snapshot should preserve task_type hints from the goal text'
+assert proposal['analysis']['evaluationProfile'] == expected_eval_profile, 'explicit-goal proposal snapshot should preserve evaluation_profile hints from the goal text'
+assert proposal['analysis']['critique'] == ['Keep critique notes separate from the mission anchor.'], 'explicit-goal proposal snapshot should preserve critique hints from the goal text'
+assert proposal['analysis']['risks'] == ['Session-only scope could leak into the next workflow round.'], 'explicit-goal proposal snapshot should preserve risk hints from the goal text'
+assert 'Critique:' not in proposal['goalText'], 'goalText should keep critique notes separate from mission/scope/constraints/acceptance'
+assert 'Task type:' not in proposal['goalText'], 'goalText should keep task_type hints separate from the mission body'
 assert state['current_phase'] == 'reground', 'current_phase should reset to reground after explicit-goal next-round start'
 assert state['continuation_policy'] == 'continue', 'continuation_policy should reset to continue after explicit-goal next-round start'
 assert state['project_done'] is False, 'project_done should reset to false after explicit-goal next-round start'
 assert state['requires_reground'] is True, 'requires_reground should reset to true after explicit-goal next-round start'
 assert state['next_mandatory_role'] == 'completion-regrounder', 'next role should reset to completion-regrounder after explicit-goal next-round start'
 assert continuation_reason.startswith('User refocused workflow via /cook:'), 'continuation_reason should record the explicit-goal next-round start'
+assert 'task_type=completion-workflow' in continuation_reason, 'explicit-goal next-round start should persist the selected task_type'
+assert 'evaluation_profile=completion-rubric-v1' in continuation_reason, 'explicit-goal next-round start should persist the selected evaluation_profile'
+assert 'Keep critique notes separate from the mission anchor.' in continuation_reason, 'explicit-goal next-round start should persist the accepted critique outcome'
 assert 'Keep explicit scope.' in continuation_reason, 'explicit scope should remain in the explicit-goal proposal'
 assert 'Add session-only scope.' not in continuation_reason, 'session-derived scope should not be merged when analyst output is unavailable'
 assert 'Restyle widget.' not in continuation_reason, 'unrelated session-derived scope should not be merged when analyst output is unavailable'
@@ -283,19 +363,33 @@ import json
 from pathlib import Path
 
 mission = 'Use a proposal analyst to summarize natural discussion before /cook writes canonical state.'
+expected_task_type = 'completion-workflow'
+expected_eval_profile = 'completion-rubric-v1'
 mission_text = Path('.agent/mission.md').read_text()
+profile = json.loads(Path('.agent/profile.json').read_text())
 state = json.loads(Path('.agent/state.json').read_text())
 plan = json.loads(Path('.agent/plan.json').read_text())
 active = json.loads(Path('.agent/active-slice.json').read_text())
 continuation_reason = state['continuation_reason']
 
 assert mission in mission_text, '.agent/mission.md did not record the analyst-derived mission anchor'
+assert profile['task_type'] == expected_task_type, 'profile.json task_type mismatch after analyst-derived restart'
+assert profile['evaluation_profile'] == expected_eval_profile, 'profile.json evaluation_profile mismatch after analyst-derived restart'
 assert state['mission_anchor'] == mission, 'state.json mission_anchor mismatch after analyst-derived bootstrap'
+assert state['task_type'] == expected_task_type, 'state.json task_type mismatch after analyst-derived bootstrap'
+assert state['evaluation_profile'] == expected_eval_profile, 'state.json evaluation_profile mismatch after analyst-derived bootstrap'
 assert plan['mission_anchor'] == mission, 'plan.json mission_anchor mismatch after analyst-derived bootstrap'
+assert plan['task_type'] == expected_task_type, 'plan.json task_type mismatch after analyst-derived bootstrap'
+assert plan['evaluation_profile'] == expected_eval_profile, 'plan.json evaluation_profile mismatch after analyst-derived bootstrap'
 assert active['mission_anchor'] == mission, 'active-slice.json mission_anchor mismatch after analyst-derived bootstrap'
+assert active['task_type'] == expected_task_type, 'active-slice.json task_type mismatch after analyst-derived bootstrap'
+assert active['evaluation_profile'] == expected_eval_profile, 'active-slice.json evaluation_profile mismatch after analyst-derived bootstrap'
 assert state['current_phase'] == 'reground', 'current_phase should reset to reground after analyst-derived bootstrap'
 assert state['next_mandatory_role'] == 'completion-regrounder', 'next role should reset to completion-regrounder after analyst-derived bootstrap'
 assert continuation_reason.startswith('User refocused workflow via /cook:'), 'continuation_reason should record the analyst-derived restart'
+assert 'task_type=completion-workflow' in continuation_reason, 'analyst-derived restart should persist the selected task_type'
+assert 'evaluation_profile=completion-rubric-v1' in continuation_reason, 'analyst-derived restart should persist the selected evaluation_profile'
+assert 'critique outcome=accepted critique=none' in continuation_reason, 'analyst-derived restart should persist that no critique notes were accepted'
 assert 'Keep explicit goals anchored.' in continuation_reason, 'analyst-derived scope should be preserved'
 PY
 
@@ -307,7 +401,7 @@ git init -q
 
 UI_SESSION_START="$TMPDIR/ui-session-start.jsonl"
 UI_DISCUSSION_START=$'Mission: Replace the crowded selector with a clearer action layout.\nScope:\n- Separate proposal text from actions.\nConstraints:\n- Preserve Start/Edit/Cancel behavior.\nAcceptance:\n- Add regression coverage.'
-UI_ANALYST_OUTPUT_START='{"mission":"Replace the crowded selector with a clearer action layout.","scope":["Separate proposal text from actions."],"constraints":["Preserve Start/Edit/Cancel behavior."],"acceptance":["Add regression coverage."],"confidence":0.95}'
+UI_ANALYST_OUTPUT_START='{"mission":"Replace the crowded selector with a clearer action layout.","scope":["Separate proposal text from actions."],"constraints":["Preserve Start/Edit/Cancel behavior."],"acceptance":["Add regression coverage."],"critique":["Keep critique details separate from the editable proposal body."],"risks":["Bundling critique into the action list would make the confirmation harder to scan."],"task_type":"completion-workflow","evaluation_profile":"completion-rubric-v1","possible_noise":["old selector wording"],"confidence":0.95}'
 UI_SNAPSHOT_START="$TMPDIR/context-proposal-ui-start.json"
 write_session "$UI_SESSION_START" "$UI_ROOT_START" "$UI_DISCUSSION_START"
 
@@ -326,13 +420,25 @@ snapshot = json.loads(Path(sys.argv[1]).read_text())
 state = json.loads(Path('.agent/state.json').read_text())
 
 assert snapshot['proposalHeading'] == 'Proposed workflow', 'custom confirmation snapshot should expose a dedicated proposal section'
+assert snapshot['critiqueHeading'] == 'Critique and risks', 'custom confirmation snapshot should expose critique separately from the proposal body'
+assert snapshot['routingHeading'] == 'Routing recommendations', 'custom confirmation snapshot should expose routing recommendations separately from the proposal body'
+assert state['task_type'] == 'completion-workflow', 'start action should preserve canonical task_type'
+assert state['evaluation_profile'] == 'completion-rubric-v1', 'start action should preserve canonical evaluation_profile'
 assert 'Mission\nReplace the crowded selector with a clearer action layout.' in snapshot['proposalBody'], 'proposal body should be captured separately from the action list'
+assert 'Keep critique details separate from the editable proposal body.' not in snapshot['proposalBody'], 'critique notes should not be embedded in the proposal body'
+assert 'Critique\n- Keep critique details separate from the editable proposal body.' in snapshot['critiqueBody'], 'critique section should render accepted critique notes separately'
+assert 'Risks\n- Bundling critique into the action list would make the confirmation harder to scan.' in snapshot['critiqueBody'], 'critique section should render risk notes separately'
+assert 'Possible noise\n- old selector wording' in snapshot['critiqueBody'], 'critique section should render possible-noise notes separately'
+assert '- task_type: completion-workflow' in snapshot['routingBody'], 'routing section should render the recommended task_type'
+assert '- evaluation_profile: completion-rubric-v1' in snapshot['routingBody'], 'routing section should render the recommended evaluation_profile'
 assert [action['id'] for action in snapshot['actions']] == ['start', 'edit', 'cancel'], 'custom confirmation actions should stay Start/Edit/Cancel'
 assert [action['label'] for action in snapshot['actions']] == ['Start', 'Edit', 'Cancel'], 'custom confirmation action labels should be concise'
 for action in snapshot['actions']:
     assert 'Replace the crowded selector with a clearer action layout.' not in action['label'], 'proposal mission should not be embedded in action labels'
     assert 'Separate proposal text from actions.' not in action['description'], 'proposal scope should not be embedded in action descriptions'
 assert state['mission_anchor'] == 'Replace the crowded selector with a clearer action layout.', 'start action should still accept the proposed mission'
+assert state['continuation_reason'].startswith('User started workflow via /cook:'), 'start action should persist the startup routing outcome in continuation_reason'
+assert 'Keep critique details separate from the editable proposal body.' in state['continuation_reason'], 'start action should persist the accepted critique outcome canonically'
 PY
 
 # Custom confirmation UI: edit should keep the existing editor/parsing flow when the action comes from the new layout.
@@ -343,7 +449,7 @@ git init -q
 
 UI_SESSION_EDIT="$TMPDIR/ui-session-edit.jsonl"
 UI_DISCUSSION_EDIT=$'Mission: Keep editing support in the custom confirmation UI.\nScope:\n- Preserve the proposal editor.\nConstraints:\n- Keep parsing structured proposal text.\nAcceptance:\n- Update the mission anchor after edit.'
-UI_ANALYST_OUTPUT_EDIT='{"mission":"Keep editing support in the custom confirmation UI.","scope":["Preserve the proposal editor."],"constraints":["Keep parsing structured proposal text."],"acceptance":["Update the mission anchor after edit."],"confidence":0.94}'
+UI_ANALYST_OUTPUT_EDIT='{"mission":"Keep editing support in the custom confirmation UI.","scope":["Preserve the proposal editor."],"constraints":["Keep parsing structured proposal text."],"acceptance":["Update the mission anchor after edit."],"critique":["Keep critique persistence even when the operator edits the proposal body."],"task_type":"completion-workflow","evaluation_profile":"completion-rubric-v1","confidence":0.94}'
 UI_EDIT_TEXT=$'Mission: Edited mission from the custom confirmation UI.\nScope:\n- Preserve parsing after edit.\nConstraints:\n- Keep the shared confirmation flow.\nAcceptance:\n- Start the workflow from the edited proposal.'
 write_session "$UI_SESSION_EDIT" "$UI_ROOT_EDIT" "$UI_DISCUSSION_EDIT"
 
@@ -363,9 +469,12 @@ active = json.loads(Path('.agent/active-slice.json').read_text())
 mission = 'Edited mission from the custom confirmation UI.'
 
 assert state['mission_anchor'] == mission, 'edit action should still route through the proposal parser and update state.json'
+assert state['task_type'] == 'completion-workflow', 'edit action should preserve canonical task_type'
+assert state['evaluation_profile'] == 'completion-rubric-v1', 'edit action should preserve canonical evaluation_profile'
 assert plan['mission_anchor'] == mission, 'edit action should still route through the proposal parser and update plan.json'
 assert active['mission_anchor'] == mission, 'edit action should still route through the proposal parser and update active-slice.json'
 assert state['current_phase'] == 'reground', 'edit action should still bootstrap/reground the workflow'
+assert 'Keep critique persistence even when the operator edits the proposal body.' in state['continuation_reason'], 'edit action should preserve the accepted critique outcome canonically'
 PY
 
 # Custom confirmation UI: cancel should exit without writing canonical state.

package/scripts/refocus-test.sh

@@ -44,15 +44,26 @@ import json
 from pathlib import Path
 
 new_anchor = 'refocused smoke-test mission with tests and docs parity.'
+expected_task_type = 'completion-workflow'
+expected_eval_profile = 'completion-rubric-v1'
 mission_text = Path('.agent/mission.md').read_text()
+profile = json.loads(Path('.agent/profile.json').read_text())
 state = json.loads(Path('.agent/state.json').read_text())
 plan = json.loads(Path('.agent/plan.json').read_text())
 active = json.loads(Path('.agent/active-slice.json').read_text())
 
 assert new_anchor in mission_text, '.agent/mission.md did not update to the refocused mission anchor'
+assert profile['task_type'] == expected_task_type, 'profile.json task_type mismatch after refocus'
+assert profile['evaluation_profile'] == expected_eval_profile, 'profile.json evaluation_profile mismatch after refocus'
 assert state['mission_anchor'] == new_anchor, 'state.json mission_anchor mismatch after refocus'
+assert state['task_type'] == expected_task_type, 'state.json task_type mismatch after refocus'
+assert state['evaluation_profile'] == expected_eval_profile, 'state.json evaluation_profile mismatch after refocus'
 assert plan['mission_anchor'] == new_anchor, 'plan.json mission_anchor mismatch after refocus'
+assert plan['task_type'] == expected_task_type, 'plan.json task_type mismatch after refocus'
+assert plan['evaluation_profile'] == expected_eval_profile, 'plan.json evaluation_profile mismatch after refocus'
 assert active['mission_anchor'] == new_anchor, 'active-slice.json mission_anchor mismatch after refocus'
+assert active['task_type'] == expected_task_type, 'active-slice.json task_type mismatch after refocus'
+assert active['evaluation_profile'] == expected_eval_profile, 'active-slice.json evaluation_profile mismatch after refocus'
 assert state['current_phase'] == 'reground', 'state.json current_phase should reset to reground after refocus'
 assert state['requires_reground'] is True, 'state.json requires_reground should be true after refocus'
 assert state['next_mandatory_role'] == 'completion-regrounder', 'next_mandatory_role should reset to completion-regrounder'

package/scripts/release-check.sh

@@ -4,10 +4,12 @@ set -euo pipefail
 ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
 cd "$ROOT"
 
+echo "[release-check] running startup/refocus/context regressions, including critique-aware /cook confirmation coverage"
 npm run smoke-test
 npm run refocus-test
 npm run context-proposal-test
 npm run observability-status-test
+npm run rubric-contract-test
 npm pack --dry-run >/dev/null
 
 echo "release check passed"

package/scripts/rubric-contract-test.sh

@@ -0,0 +1,249 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
+cd "$ROOT"
+
+node <<'NODE'
+const fs = require('node:fs');
+
+const read = (file) => fs.readFileSync(file, 'utf8');
+const assertIncludes = (file, snippet) => {
+  const text = read(file);
+  if (!text.includes(snippet)) {
+    console.error(`${file} is missing required rubric-contract text: ${snippet}`);
+    process.exit(1);
+  }
+};
+
+const rubricHeading = '## Structured Evaluation Rubric Foundation';
+const rubricDimensions = [
+  'Contract coverage',
+  'Correctness risk',
+  'Verification evidence',
+  'Docs/state parity',
+];
+const verdictSnippets = [
+  '`pass` — no material issue remains',
+  '`concern` — a real caveat or remaining gap exists',
+  '`fail` — a blocking issue or contradictory truth exists',
+];
+
+for (const file of [
+  'skills/completion-protocol/SKILL.md',
+  'skills/completion-protocol/references/completion.md',
+]) {
+  assertIncludes(file, rubricHeading);
+  assertIncludes(file, 'canonical `task_type` and `evaluation_profile` signaling');
+  assertIncludes(file, 'routing metadata only; later slices may still add stricter profile-aware rubric-output enforcement');
+  assertIncludes(file, '- `Rubric:`');
+  for (const dimension of rubricDimensions) {
+    assertIncludes(file, `- \`- ${dimension}: pass|concern|fail - ...\``);
+  }
+  for (const snippet of verdictSnippets) {
+    assertIncludes(file, snippet);
+  }
+}
+
+for (const file of [
+  'agents/completion-reviewer.md',
+  'agents/completion-auditor.md',
+  'agents/completion-stop-judge.md',
+]) {
+  assertIncludes(file, 'Always emit the shared rubric section');
+  assertIncludes(file, 'Use these exact rubric dimension names and verdict words');
+  assertIncludes(file, '`evaluation_profile`');
+  assertIncludes(file, '`implementation_surfaces`');
+  assertIncludes(file, '`verification_commands`');
+  assertIncludes(file, '- `Rubric:`');
+  for (const dimension of rubricDimensions) {
+    assertIncludes(file, `- \`- ${dimension}: pass|concern|fail - ...\``);
+  }
+}
+
+assertIncludes('README.md', '## Structured evaluation rubrics');
+assertIncludes('README.md', '- `task_type: completion-workflow`');
+assertIncludes('README.md', '- `evaluation_profile: completion-rubric-v1`');
+assertIncludes('README.md', 'kickoff/reminder/resume text and reviewer/auditor/stop-judge evaluation handoffs so downstream roles can rely on canonical signaling instead of prose inference alone.');
+assertIncludes('README.md', 'Reviewer, auditor, and stop-judge dispatch/reminder surfaces now also thread the current active-slice implementation contract');
+assertIncludes('README.md', 'Canonical reviewer/auditor/stop-judge transcription now fails closed on malformed rubric-bearing reports');
+assertIncludes('README.md', 'npm run rubric-contract-test`, which now exercises reviewer, auditor, and stop-judge transcription paths');
+for (const dimension of rubricDimensions) {
+  assertIncludes('README.md', `- \`${dimension}\``);
+}
+
+assertIncludes('CHANGELOG.md', 'shared structured evaluation-rubric contract');
+assertIncludes('CHANGELOG.md', 'added canonical `task_type: completion-workflow` and `evaluation_profile: completion-rubric-v1` signaling across the packaged control-plane defaults, verifier schema, and kickoff/reminder/resume surfaces');
+assertIncludes('CHANGELOG.md', 'threaded canonical `evaluation_profile` plus the active-slice implementation contract into reviewer/auditor/stop-judge reminder and dispatch surfaces');
+assertIncludes('CHANGELOG.md', 'made reviewer/auditor/stop-judge transcription fail closed on malformed rubric-bearing outputs while still accepting valid reports');
+assertIncludes('extensions/completion/index.ts', 'Canonical routing profile:\\n- task_type: ${taskType}\\n- evaluation_profile: ${evaluationProfile}');
+assertIncludes('extensions/completion/index.ts', '`Task type: ${currentTaskType(snapshot) ?? "(missing)"}`');
+assertIncludes('extensions/completion/index.ts', '`Evaluation profile: ${currentEvaluationProfile(snapshot) ?? "(missing)"}`');
+assertIncludes('extensions/completion/index.ts', '`task_type: ${currentTaskType(snapshot) ?? "(missing)"}`');
+assertIncludes('extensions/completion/index.ts', '`evaluation_profile: ${currentEvaluationProfile(snapshot) ?? "(missing)"}`');
+assertIncludes('extensions/completion/index.ts', 'Canonical evaluation handoff for ${role}:');
+assertIncludes('extensions/completion/index.ts', 'buildEvaluationRoleReminderText(snapshot, nextRole)');
+assertIncludes('extensions/completion/index.ts', 'roleReporting.transcribeCanonicalRoleReport');
+assertIncludes('extensions/completion/role-reporting.js', 'Missing Rubric heading for ${role}.');
+assertIncludes('extensions/completion/role-reporting.js', 'Reviewer output cannot mark \'Acceptable as-is: yes\' when any rubric line is fail.');
+assertIncludes('extensions/completion/role-reporting.js', 'Auditor output must answer \'Tracked and unignored worktree is clean\' with yes or no.');
+assertIncludes('extensions/completion/role-reporting.js', 'Auditor output must answer \'Stale or conflicting canonical state\' with yes or no.');
+assertIncludes('extensions/completion/role-reporting.js', 'Auditor output must answer \'Plan truthfully captures remaining slice backlog\' with yes or no.');
+assertIncludes('extensions/completion/role-reporting.js', 'Stop-judge output cannot mark \'Can the project stop now: yes\' when any rubric line is fail.');
+assertIncludes('extensions/completion/role-reporting.js', 'Stop-judge output must answer \'Docs/config/runbooks match shipped behavior\' with yes or no.');
+assertIncludes('extensions/completion/role-reporting.js', 'Stop-judge output must answer \'Tracked and unignored worktree is clean\' with yes or no.');
+assertIncludes('package.json', '"rubric-contract-test": "bash ./scripts/rubric-contract-test.sh"');
+assertIncludes('scripts/release-check.sh', 'npm run rubric-contract-test');
+assertIncludes('.agent/verify_completion_stop.sh', 'npm run release-check >/dev/null');
+NODE
+
+node <<'NODE'
+const fs = require('node:fs');
+const path = require('node:path');
+const {
+  parseReportFields,
+  transcribeCanonicalRoleReport,
+} = require('./extensions/completion/role-reporting.js');
+
+const tempRootBase = path.join(process.cwd(), '.agent', 'tmp');
+fs.mkdirSync(tempRootBase, { recursive: true });
+const tempRoot = fs.mkdtempSync(path.join(tempRootBase, 'rubric-role-reporting-'));
+const snapshotFiles = {
+  sliceHistoryPath: path.join(tempRoot, 'slice-history.jsonl'),
+  stopHistoryPath: path.join(tempRoot, 'stop-check-history.jsonl'),
+};
+fs.writeFileSync(snapshotFiles.sliceHistoryPath, '');
+fs.writeFileSync(snapshotFiles.stopHistoryPath, '');
+
+const readJsonl = (file) => fs.readFileSync(file, 'utf8').split('\n').filter(Boolean).map((line) => JSON.parse(line));
+const assert = (condition, message) => {
+  if (!condition) throw new Error(message);
+};
+
+const reviewerReport = `MISSION ANCHOR: test mission\nRemaining contract IDs: TEST-CONTRACT\nRubric:\n- Contract coverage: pass - Locked acceptance criteria match the committed slice.\n- Correctness risk: pass - No blocking regression is evident.\n- Verification evidence: pass - Deterministic proof was rerun successfully.\n- Docs/state parity: pass - Docs and canonical state are aligned.\nFindings: none.\nAcceptable as-is: yes\nSmallest follow-up slice: none.`;
+
+const reviewerMalformed = `MISSION ANCHOR: test mission\nRemaining contract IDs: TEST-CONTRACT\nRubric:\n- Contract coverage: pass - Locked acceptance criteria match the committed slice.\n- Correctness risk: pass - No blocking regression is evident.\n- Verification evidence: pass - Deterministic proof was rerun successfully.\nFindings: none.\nAcceptable as-is: yes\nSmallest follow-up slice: none.`;
+
+const auditorReport = `MISSION ANCHOR: test mission\nRemaining contract IDs: TEST-CONTRACT\nRubric:\n- Contract coverage: pass - The accepted slice remains satisfied on HEAD.\n- Correctness risk: concern - Remaining planned work still keeps the project open.\n- Verification evidence: pass - Verification was rerun for the accepted slice.\n- Docs/state parity: pass - Canonical state can be reconciled truthfully.\nWhy the project is still not done: One planned contract remains after this accepted slice.\nOpen top-level contract IDs: TEST-CONTRACT\nBlocker count: 0\nHigh-value gap count: 1\nTracked and unignored worktree is clean: yes\nWorktree blockers: none\nNext mandatory slice: next-slice\nStale or conflicting canonical state: no\nPlan truthfully captures remaining slice backlog: yes - one planned slice remains.`;
+
+const auditorMalformed = `MISSION ANCHOR: test mission\nRemaining contract IDs: TEST-CONTRACT\nWhy the project is still not done: One planned contract remains after this accepted slice.\nOpen top-level contract IDs: TEST-CONTRACT\nBlocker count: 0\nHigh-value gap count: 1\nTracked and unignored worktree is clean: yes\nWorktree blockers: none\nNext mandatory slice: next-slice\nStale or conflicting canonical state: no\nPlan truthfully captures remaining slice backlog: yes - one planned slice remains.`;
+
+const stopJudgeReport = `MISSION ANCHOR: test mission\nRemaining contract IDs: none\nRubric:\n- Contract coverage: pass - All implementation slices are accepted on HEAD.\n- Correctness risk: pass - No remaining blocker or high-value gap is evident.\n- Verification evidence: pass - Final verification passes for the current head.\n- Docs/state parity: pass - Docs, config, and canonical state match shipped behavior.\nCan the project stop now: yes\nExact remaining open top-level contract IDs: none\nBlocker count: 0\nHigh-value gap count: 0\nLatest completed slice commit: abcdef1234567890abcdef1234567890abcdef12\nDocs/config/runbooks match shipped behavior: yes\nTracked and unignored worktree is clean: yes\nBrief justification: Current HEAD satisfies the stop criteria.`;
+
+const stopJudgeMalformed = `MISSION ANCHOR: test mission\nRemaining contract IDs: none\nRubric:\n- Contract coverage: fail - A blocking contract is still open.\n- Correctness risk: pass - No additional risk was found.\n- Verification evidence: pass - Verification still passes.\n- Docs/state parity: pass - Docs and state match.\nCan the project stop now: yes\nExact remaining open top-level contract IDs: TEST-CONTRACT\nBlocker count: 1\nHigh-value gap count: 0\nLatest completed slice commit: abcdef1234567890abcdef1234567890abcdef12\nDocs/config/runbooks match shipped behavior: yes\nTracked and unignored worktree is clean: yes\nBrief justification: This should be rejected because the rubric blocks stop.`;
+
+const auditorMalformedYesNo = `MISSION ANCHOR: test mission\nRemaining contract IDs: TEST-CONTRACT\nRubric:\n- Contract coverage: pass - The accepted slice remains satisfied on HEAD.\n- Correctness risk: concern - Remaining planned work still keeps the project open.\n- Verification evidence: pass - Verification was rerun for the accepted slice.\n- Docs/state parity: pass - Canonical state can be reconciled truthfully.\nWhy the project is still not done: One planned contract remains after this accepted slice.\nOpen top-level contract IDs: TEST-CONTRACT\nBlocker count: 0\nHigh-value gap count: 1\nTracked and unignored worktree is clean: maybe\nWorktree blockers: none\nNext mandatory slice: next-slice\nStale or conflicting canonical state: perhaps\nPlan truthfully captures remaining slice backlog: sorta.`;
+
+const stopJudgeMalformedYesNo = `MISSION ANCHOR: test mission\nRemaining contract IDs: none\nRubric:\n- Contract coverage: pass - All implementation slices are accepted on HEAD.\n- Correctness risk: pass - No remaining blocker or high-value gap is evident.\n- Verification evidence: pass - Final verification passes for the current head.\n- Docs/state parity: pass - Docs, config, and canonical state match shipped behavior.\nCan the project stop now: no\nExact remaining open top-level contract IDs: TEST-CONTRACT\nBlocker count: 0\nHigh-value gap count: 1\nLatest completed slice commit: abcdef1234567890abcdef1234567890abcdef12\nDocs/config/runbooks match shipped behavior: maybe\nTracked and unignored worktree is clean: perhaps\nBrief justification: This should be rejected because malformed yes/no-style fields must fail closed.`;
+
+(async () => {
+  const reviewed = await transcribeCanonicalRoleReport({
+    role: 'completion-reviewer',
+    output: reviewerReport,
+    reportFields: parseReportFields(reviewerReport),
+    snapshotFiles,
+    headSha: '1111111111111111111111111111111111111111',
+    sliceId: 'slice-review',
+    recordedAt: 1,
+  });
+  assert(reviewed.errors.length === 0, `reviewer valid report should transcribe cleanly: ${reviewed.errors.join(' | ')}`);
+  assert(reviewed.appended.includes('reviewed:slice-review'), 'reviewer transcription should append reviewed record');
+  assert(readJsonl(snapshotFiles.sliceHistoryPath).length === 1, 'reviewer transcription should create one slice-history record');
+
+  const reviewerRejected = await transcribeCanonicalRoleReport({
+    role: 'completion-reviewer',
+    output: reviewerMalformed,
+    reportFields: parseReportFields(reviewerMalformed),
+    snapshotFiles,
+    headSha: '2222222222222222222222222222222222222222',
+    sliceId: 'slice-review',
+    recordedAt: 2,
+  });
+  assert(reviewerRejected.errors.some((error) => error.includes('Docs/state parity')), 'reviewer malformed report should be rejected for missing rubric line');
+  assert(readJsonl(snapshotFiles.sliceHistoryPath).length === 1, 'rejected reviewer report must not append history');
+
+  const audited = await transcribeCanonicalRoleReport({
+    role: 'completion-auditor',
+    output: auditorReport,
+    reportFields: parseReportFields(auditorReport),
+    snapshotFiles,
+    headSha: '3333333333333333333333333333333333333333',
+    sliceId: 'slice-audit',
+    recordedAt: 3,
+  });
+  assert(audited.errors.length === 0, `auditor valid report should transcribe cleanly: ${audited.errors.join(' | ')}`);
+  assert(audited.appended.includes('audited:slice-audit'), 'auditor transcription should append audited record');
+  assert(readJsonl(snapshotFiles.sliceHistoryPath).length === 2, 'auditor transcription should append a second slice-history record');
+
+  const auditorRejected = await transcribeCanonicalRoleReport({
+    role: 'completion-auditor',
+    output: auditorMalformed,
+    reportFields: parseReportFields(auditorMalformed),
+    snapshotFiles,
+    headSha: '4444444444444444444444444444444444444444',
+    sliceId: 'slice-audit',
+    recordedAt: 4,
+  });
+  assert(auditorRejected.errors.some((error) => error.includes('Missing Rubric heading')), 'auditor malformed report should be rejected without rubric heading');
+  assert(readJsonl(snapshotFiles.sliceHistoryPath).length === 2, 'rejected auditor report must not append history');
+
+  const judged = await transcribeCanonicalRoleReport({
+    role: 'completion-stop-judge',
+    output: stopJudgeReport,
+    reportFields: parseReportFields(stopJudgeReport),
+    snapshotFiles,
+    headSha: '5555555555555555555555555555555555555555',
+    recordedAt: 5,
+  });
+  assert(judged.errors.length === 0, `stop-judge valid report should transcribe cleanly: ${judged.errors.join(' | ')}`);
+  assert(judged.appended.includes('judgment:555555555555'), 'stop-judge transcription should append judgment record');
+  assert(readJsonl(snapshotFiles.stopHistoryPath).length === 1, 'stop-judge transcription should create one judgment record');
+
+  const judgeRejected = await transcribeCanonicalRoleReport({
+    role: 'completion-stop-judge',
+    output: stopJudgeMalformed,
+    reportFields: parseReportFields(stopJudgeMalformed),
+    snapshotFiles,
+    headSha: '6666666666666666666666666666666666666666',
+    recordedAt: 6,
+  });
+  assert(judgeRejected.errors.some((error) => error.includes("Can the project stop now: yes")), 'stop-judge malformed report should be rejected when fail rubric contradicts yes verdict');
+  assert(readJsonl(snapshotFiles.stopHistoryPath).length === 1, 'rejected stop-judge report must not append judgment history');
+
+  const auditorYesNoRejected = await transcribeCanonicalRoleReport({
+    role: 'completion-auditor',
+    output: auditorMalformedYesNo,
+    reportFields: parseReportFields(auditorMalformedYesNo),
+    snapshotFiles,
+    headSha: '7777777777777777777777777777777777777777',
+    sliceId: 'slice-audit',
+    recordedAt: 7,
+  });
+  assert(auditorYesNoRejected.errors.some((error) => error.includes("Tracked and unignored worktree is clean")), 'auditor malformed yes/no report should reject invalid worktree cleanliness values');
+  assert(auditorYesNoRejected.errors.some((error) => error.includes("Stale or conflicting canonical state")), 'auditor malformed yes/no report should reject invalid canonical-state values');
+  assert(auditorYesNoRejected.errors.some((error) => error.includes("Plan truthfully captures remaining slice backlog")), 'auditor malformed yes/no report should reject invalid backlog-truth values');
+  assert(readJsonl(snapshotFiles.sliceHistoryPath).length === 2, 'rejected auditor yes/no report must not append history');
+
+  const stopJudgeYesNoRejected = await transcribeCanonicalRoleReport({
+    role: 'completion-stop-judge',
+    output: stopJudgeMalformedYesNo,
+    reportFields: parseReportFields(stopJudgeMalformedYesNo),
+    snapshotFiles,
+    headSha: '8888888888888888888888888888888888888888',
+    recordedAt: 8,
+  });
+  assert(stopJudgeYesNoRejected.errors.some((error) => error.includes("Docs/config/runbooks match shipped behavior")), 'stop-judge malformed yes/no report should reject invalid docs parity values');
+  assert(stopJudgeYesNoRejected.errors.some((error) => error.includes("Tracked and unignored worktree is clean")), 'stop-judge malformed yes/no report should reject invalid worktree cleanliness values');
+  assert(readJsonl(snapshotFiles.stopHistoryPath).length === 1, 'rejected stop-judge yes/no report must not append judgment history');
+
+  fs.rmSync(tempRoot, { recursive: true, force: true });
+})().catch((error) => {
+  try {
+    fs.rmSync(tempRoot, { recursive: true, force: true });
+  } catch {}
+  console.error(error instanceof Error ? error.message : String(error));
+  process.exit(1);
+});
+NODE
+
+echo "rubric contract test passed"