@linimin/pi-letscook 0.1.29 → 0.1.31

package/scripts/smoke-test.sh

@@ -5,31 +5,122 @@ PKG_ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
 TMPDIR="$(mktemp -d)"
 trap 'rm -rf "$TMPDIR"' EXIT
 
-cd "$TMPDIR"
+ROOT="$TMPDIR/repo"
+KICKOFF_PROMPT="$TMPDIR/kickoff-prompt.txt"
+RESUME_PROMPT="$TMPDIR/resume-prompt.txt"
+
+mkdir -p "$ROOT"
+cd "$ROOT"
 git init -q
 
-pi -e "$PKG_ROOT" -p "/cook smoke-test mission" >/tmp/pi-completion-smoke.out 2>/tmp/pi-completion-smoke.err &
-PI_PID=$!
-for _ in $(seq 1 60); do
-  if [[ -f .agent/profile.json && -f .agent/state.json && -f .agent/plan.json && -f .agent/active-slice.json ]]; then
-    break
-  fi
-  sleep 1
+PI_COMPLETION_SKIP_DRIVER_KICKOFF=1 \
+PI_COMPLETION_TEST_DRIVER_PROMPT_PATH="$KICKOFF_PROMPT" \
+pi -e "$PKG_ROOT" -p "/cook smoke-test mission" \
+  >"$TMPDIR/pi-completion-smoke-bootstrap.out" 2>"$TMPDIR/pi-completion-smoke-bootstrap.err"
+
+for file in .agent/profile.json .agent/state.json .agent/plan.json .agent/active-slice.json; do
+  [[ -f "$file" ]] || { echo "missing canonical bootstrap file: $file" >&2; exit 1; }
 done
-if [[ ! -f .agent/profile.json || ! -f .agent/state.json || ! -f .agent/plan.json || ! -f .agent/active-slice.json ]]; then
-  echo "completion bootstrap did not materialize canonical files in time" >&2
-  cat /tmp/pi-completion-smoke.err >&2 || true
-  kill "$PI_PID" >/dev/null 2>&1 || true
-  wait "$PI_PID" >/dev/null 2>&1 || true
-  exit 1
-fi
-kill "$PI_PID" >/dev/null 2>&1 || true
-wait "$PI_PID" >/dev/null 2>&1 || true
 
 bash .agent/verify_completion_control_plane.sh >/dev/null
 bash .agent/verify_completion_stop.sh >/dev/null
 
-python3 - <<'PY2'
+python3 - "$KICKOFF_PROMPT" <<'PY'
+import json
+import sys
+from pathlib import Path
+
+expected_task_type = 'completion-workflow'
+expected_eval_profile = 'completion-rubric-v1'
+
+profile = json.loads(Path('.agent/profile.json').read_text())
+state = json.loads(Path('.agent/state.json').read_text())
+plan = json.loads(Path('.agent/plan.json').read_text())
+active = json.loads(Path('.agent/active-slice.json').read_text())
+kickoff = Path(sys.argv[1]).read_text()
+
+assert profile['task_type'] == expected_task_type, 'profile.json task_type mismatch after bootstrap'
+assert profile['evaluation_profile'] == expected_eval_profile, 'profile.json evaluation_profile mismatch after bootstrap'
+assert state['task_type'] == expected_task_type, 'state.json task_type mismatch after bootstrap'
+assert state['evaluation_profile'] == expected_eval_profile, 'state.json evaluation_profile mismatch after bootstrap'
+assert plan['task_type'] == expected_task_type, 'plan.json task_type mismatch after bootstrap'
+assert plan['evaluation_profile'] == expected_eval_profile, 'plan.json evaluation_profile mismatch after bootstrap'
+assert active['task_type'] == expected_task_type, 'active-slice.json task_type mismatch after bootstrap'
+assert active['evaluation_profile'] == expected_eval_profile, 'active-slice.json evaluation_profile mismatch after bootstrap'
+assert active['implementation_surfaces'] == [], 'active-slice.json should scaffold empty implementation_surfaces'
+assert active['verification_commands'] == [], 'active-slice.json should scaffold empty verification_commands'
+assert 'Canonical routing profile:' in kickoff, 'kickoff prompt should expose canonical routing profile'
+assert f'- task_type: {expected_task_type}' in kickoff, 'kickoff prompt missing canonical task_type'
+assert f'- evaluation_profile: {expected_eval_profile}' in kickoff, 'kickoff prompt missing canonical evaluation_profile'
+PY
+
+PI_COMPLETION_SKIP_DRIVER_KICKOFF=1 \
+PI_COMPLETION_TEST_DRIVER_PROMPT_PATH="$RESUME_PROMPT" \
+pi -e "$PKG_ROOT" -p "/cook" \
+  >"$TMPDIR/pi-completion-smoke-resume.out" 2>"$TMPDIR/pi-completion-smoke-resume.err"
+
+python3 - "$RESUME_PROMPT" <<'PY'
+import sys
+from pathlib import Path
+
+expected_task_type = 'completion-workflow'
+expected_eval_profile = 'completion-rubric-v1'
+resume = Path(sys.argv[1]).read_text()
+
+assert 'Canonical routing profile:' in resume, 'resume prompt should expose canonical routing profile'
+assert f'- task_type: {expected_task_type}' in resume, 'resume prompt missing canonical task_type'
+assert f'- evaluation_profile: {expected_eval_profile}' in resume, 'resume prompt missing canonical evaluation_profile'
+PY
+
+python3 - <<'PY'
+import json
+from pathlib import Path
+path = Path('.agent/state.json')
+state = json.loads(path.read_text())
+state.pop('task_type', None)
+path.write_text(json.dumps(state, indent=2) + '\n')
+PY
+
+if bash .agent/verify_completion_control_plane.sh >/dev/null 2>&1; then
+  echo "expected control-plane verification to fail when state.json omits task_type" >&2
+  exit 1
+fi
+
+python3 - <<'PY'
+import json
+from pathlib import Path
+profile = json.loads(Path('.agent/profile.json').read_text())
+state_path = Path('.agent/state.json')
+state = json.loads(state_path.read_text())
+state['task_type'] = profile['task_type']
+state_path.write_text(json.dumps(state, indent=2) + '\n')
+PY
+
+python3 - <<'PY'
+import json
+from pathlib import Path
+path = Path('.agent/active-slice.json')
+active = json.loads(path.read_text())
+active.pop('evaluation_profile', None)
+path.write_text(json.dumps(active, indent=2) + '\n')
+PY
+
+if bash .agent/verify_completion_control_plane.sh >/dev/null 2>&1; then
+  echo "expected control-plane verification to fail when active-slice.json omits evaluation_profile" >&2
+  exit 1
+fi
+
+python3 - <<'PY'
+import json
+from pathlib import Path
+profile = json.loads(Path('.agent/profile.json').read_text())
+active_path = Path('.agent/active-slice.json')
+active = json.loads(active_path.read_text())
+active['evaluation_profile'] = profile['evaluation_profile']
+active_path.write_text(json.dumps(active, indent=2) + '\n')
+PY
+
+python3 - <<'PY'
 import json
 from pathlib import Path
 path = Path('.agent/active-slice.json')
@@ -41,8 +132,10 @@ active.update({
     'contract_ids': ['smoke-contract'],
     'acceptance_criteria': ['criterion'],
     'blocked_on': [],
-    'locked_notes': [],
+    'locked_notes': ['keep the change scoped to the selected active-slice contract'],
     'must_fix_findings': [],
+    'implementation_surfaces': ['extensions/completion/index.ts', '.agent/verify_completion_control_plane.sh'],
+    'verification_commands': ['bash .agent/verify_completion_control_plane.sh', 'npm run smoke-test'],
     'basis_commit': 'deadbeef',
     'remaining_contract_ids_before': ['smoke-contract'],
     'release_blocker_count_before': 1,
@@ -51,14 +144,14 @@ active.update({
 active.pop('priority', None)
 active.pop('why_now', None)
 path.write_text(json.dumps(active, indent=2) + '\n')
-PY2
+PY
 
 if bash .agent/verify_completion_control_plane.sh >/dev/null 2>&1; then
   echo "expected control-plane verification to fail when selected active-slice omits priority/why_now" >&2
   exit 1
 fi
 
-python3 - <<'PY3'
+python3 - <<'PY'
 import json
 from pathlib import Path
 path = Path('.agent/active-slice.json')
@@ -66,9 +159,47 @@ active = json.loads(path.read_text())
 active['priority'] = 1
 active['why_now'] = 'smoke test exact handoff'
 path.write_text(json.dumps(active, indent=2) + '\n')
-PY3
+PY
+
+python3 - <<'PY'
+import json
+from pathlib import Path
+path = Path('.agent/active-slice.json')
+active = json.loads(path.read_text())
+active.pop('implementation_surfaces', None)
+active.pop('verification_commands', None)
+path.write_text(json.dumps(active, indent=2) + '\n')
+PY
+
+if bash .agent/verify_completion_control_plane.sh >/dev/null 2>&1; then
+  echo "expected control-plane verification to fail when selected active-slice omits implementation_surfaces/verification_commands" >&2
+  exit 1
+fi
+
+python3 - <<'PY'
+import json
+from pathlib import Path
+path = Path('.agent/active-slice.json')
+active = json.loads(path.read_text())
+active['implementation_surfaces'] = ['extensions/completion/index.ts', '.agent/verify_completion_control_plane.sh']
+active['verification_commands'] = ['bash .agent/verify_completion_control_plane.sh', 'npm run smoke-test']
+path.write_text(json.dumps(active, indent=2) + '\n')
+PY
 
 bash .agent/verify_completion_control_plane.sh >/dev/null
 bash .agent/verify_completion_stop.sh >/dev/null
 
-echo "smoke test passed: $TMPDIR"
+python3 - "$PKG_ROOT" <<'PY'
+import sys
+from pathlib import Path
+
+text = Path(sys.argv[1], 'extensions/completion', 'index.ts').read_text()
+assert 'Active slice priority: ${activePriority}' in text, 'system reminder source should expose active-slice priority'
+assert 'Active slice why_now: ${activeWhyNow}' in text, 'system reminder source should expose active-slice why_now'
+assert 'Active implementation surfaces: ${implementationSurfaces.join(", ")}' in text, 'system reminder source should expose implementation_surfaces'
+assert 'Active verification commands: ${verificationCommands.join(" | ")}' in text, 'system reminder source should expose verification_commands'
+assert '`- implementation_surfaces: ${implementationSurfaces.join(" | ")}`' in text, 'resume capsule source should expose implementation_surfaces'
+assert '`- verification_commands: ${verificationCommands.join(" | ")}`' in text, 'resume capsule source should expose verification_commands'
+PY
+
+echo "smoke test passed: $ROOT"

package/skills/completion-protocol/SKILL.md

@@ -148,6 +148,8 @@ The workflow driver must invoke `completion-regrounder` before continuing whenev
 - acceptance criteria for the selected or active slice are missing or unclear
 - the exact implementer handoff snapshot in `.agent/active-slice.json` is missing, stale, or contradictory
 
+The exact implementer handoff now includes implementation-scope surfaces and expected verification commands in addition to the locked slice goal, acceptance, notes, and before-slice counters.
+
 The workflow driver must not continue implementation, review, audit, or stop evaluation from compacted conversation memory alone.
 
 After compaction or recovery, `completion-implementer` must also re-read canonical `.agent/state.json`, `.agent/plan.json`, and `.agent/active-slice.json` before resuming work. If `.agent/active-slice.json` still contains a truthful exact handoff snapshot, continue from canonical state rather than asking the user to resend the original caller payload.
@@ -161,6 +163,43 @@ All completion reports must begin with:
 
 If a role-specific fixed format uses before/after wording, keep the same mission-anchor first line and then follow that role's exact format.
 
+## Structured Evaluation Rubric Foundation
+
+`completion-reviewer`, `completion-auditor`, and `completion-stop-judge` must emit rubric-backed evaluations using the same shared dimension names and verdict semantics.
+
+The shared rubric foundation now sits alongside canonical `task_type` and `evaluation_profile` signaling in the control plane. That signaling is routing metadata only; later slices may still add stricter profile-aware rubric-output enforcement.
+
+Required rubric section:
+
+- `Rubric:`
+- `- Contract coverage: pass|concern|fail - ...`
+- `- Correctness risk: pass|concern|fail - ...`
+- `- Verification evidence: pass|concern|fail - ...`
+- `- Docs/state parity: pass|concern|fail - ...`
+
+Use the dimension names and verdict words exactly as written above.
+
+Dimension meaning:
+
+- `Contract coverage` — whether the slice or current HEAD satisfies the locked acceptance criteria and role-specific workflow obligations.
+- `Correctness risk` — whether regressions, blocking defects, safety issues, or closure risks are still evident in current repo truth.
+- `Verification evidence` — whether tests, deterministic proof, and rerun verification are strong enough for the role's decision.
+- `Docs/state parity` — whether docs, config, runbooks, and canonical `.agent` state stay truthful to shipped behavior for the role's scope.
+
+Verdict semantics:
+
+- `pass` — no material issue remains for that dimension in the role's current decision.
+- `concern` — a real caveat or remaining gap exists, but it does not by itself force rejection or `NO-STOP`; explain the follow-up plainly.
+- `fail` — a blocking issue or contradictory truth exists and the role's final verdict must not be positive.
+
+Decision alignment rules:
+
+- Reviewer: any `fail` means `Acceptable as-is` must be `no`.
+- Auditor: use `concern` or `fail` to explain why the project is not yet done and whether canonical backlog/state remain truthful.
+- Stop judge: any `fail` means `Can the project stop now` must be `no`.
+
+Always include all four rubric lines, even when every dimension is `pass`.
+
 ## References
 
 Read these bundled references when you need the full protocol or scaffolding material:

package/skills/completion-protocol/references/completion.md

@@ -36,6 +36,8 @@
   "project_name": "<repo-name>",
   "required_stop_judges": 3,
   "priority_policy_id": "completion-default",
+  "task_type": "completion-workflow",
+  "evaluation_profile": "completion-rubric-v1",
   "docs_surfaces": ["README.md", "docs/"]
 }
 ```
@@ -48,6 +50,8 @@ Required fields:
 
 - `schema_version`
 - `mission_anchor`
+- `task_type`
+- `evaluation_profile`
 - `current_phase`
 - `continuation_policy`
 - `continuation_reason`
@@ -111,6 +115,8 @@ Required fields:
 
 - `schema_version`
 - `mission_anchor`
+- `task_type`
+- `evaluation_profile`
 - `last_reground_at`
 - `plan_basis`
 - `candidate_slices`
@@ -143,6 +149,27 @@ Rules:
 
 `active-slice.json` carries one current slice cursor.
 
+Required base fields:
+
+- `schema_version`
+- `mission_anchor`
+- `task_type`
+- `evaluation_profile`
+- `status`
+- `slice_id`
+- `goal`
+- `contract_ids`
+- `acceptance_criteria`
+- `blocked_on`
+- `locked_notes`
+- `must_fix_findings`
+- `implementation_surfaces`
+- `verification_commands`
+- `basis_commit`
+- `remaining_contract_ids_before`
+- `release_blocker_count_before`
+- `high_value_gap_count_before`
+
 When `status` is `selected`, `in_progress`, `committed`, or `done`, `active-slice.json` must also carry the exact implementer handoff snapshot so `completion-implementer` can resume after compaction without asking the user to resend the original caller payload.
 
 Required exact handoff fields:
@@ -153,11 +180,18 @@ Required exact handoff fields:
 - `blocked_on`
 - `locked_notes`
 - `must_fix_findings`
+- `implementation_surfaces`
+- `verification_commands`
 - `basis_commit`
 - `remaining_contract_ids_before`
 - `release_blocker_count_before`
 - `high_value_gap_count_before`
 
+Field meaning:
+
+- `implementation_surfaces` — the repo files or surfaces this slice is expected to update or keep in parity, so implementers can resume on the right scope after compaction.
+- `verification_commands` — the focused and broader deterministic checks expected before the slice is committed.
+
 Allowed `status` values:
 
 - `idle`
@@ -199,6 +233,43 @@ Minimum record shape:
 
 Empty history files are legal.
 
+## Structured Evaluation Rubric Foundation
+
+`completion-reviewer`, `completion-auditor`, and `completion-stop-judge` must emit rubric-backed evaluations using the same shared dimension names and verdict semantics.
+
+The shared rubric foundation now sits alongside canonical `task_type` and `evaluation_profile` signaling in `.agent/profile.json`, `.agent/state.json`, `.agent/plan.json`, and `.agent/active-slice.json`. That signaling is routing metadata only; later slices may still add stricter profile-aware rubric-output enforcement.
+
+Required rubric section:
+
+- `Rubric:`
+- `- Contract coverage: pass|concern|fail - ...`
+- `- Correctness risk: pass|concern|fail - ...`
+- `- Verification evidence: pass|concern|fail - ...`
+- `- Docs/state parity: pass|concern|fail - ...`
+
+Use the dimension names and verdict words exactly as written above.
+
+Dimension meaning:
+
+- `Contract coverage` — whether the slice or current HEAD satisfies the locked acceptance criteria and role-specific workflow obligations.
+- `Correctness risk` — whether regressions, blocking defects, safety issues, or closure risks are still evident in current repo truth.
+- `Verification evidence` — whether tests, deterministic proof, and rerun verification are strong enough for the role's decision.
+- `Docs/state parity` — whether docs, config, runbooks, and canonical `.agent` state stay truthful to shipped behavior for the role's scope.
+
+Verdict semantics:
+
+- `pass` — no material issue remains for that dimension in the role's current decision.
+- `concern` — a real caveat or remaining gap exists, but it does not by itself force rejection or `NO-STOP`; explain the follow-up plainly.
+- `fail` — a blocking issue or contradictory truth exists and the role's final verdict must not be positive.
+
+Decision alignment rules:
+
+- Reviewer: any `fail` means `Acceptable as-is` must be `no`.
+- Auditor: use `concern` or `fail` to explain why the project is not yet done and whether canonical backlog/state remain truthful.
+- Stop judge: any `fail` means `Can the project stop now` must be `no`.
+
+Always include all four rubric lines, even when every dimension is `pass`.
+
 ## One-Slice Lifecycle
 
 1. Re-ground from current repo truth.