pi-crew 0.11.25 → 0.11.26

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "pi-crew",
-	"version": "0.11.25",
+	"version": "0.11.26",
 	"description": "Pi extension for coordinated AI teams, workflows, worktrees, and async task orchestration",
 	"author": "baphuongna",
 	"license": "MIT",

package/src/runtime/goal-workflow/plan-templates.ts

@@ -141,7 +141,7 @@ registerPlanTemplate({
 			name: "verify",
 			role: "verifier",
 			taskTemplate:
-				"Verify that all review findings are addressed. Run FAST checks (completes in <2 min): `npm run test:critical && npx tsc --noEmit`. Do NOT run `npm run test:unit` or `npm test` — too slow (642 files, >4 min). Confirm: {{goal}} is achieved.",
+				"Verify that all review findings are addressed. Run FAST checks (completes in <2 min): `npm run test:critical && npx tsc --noEmit`. Do NOT run `npm run test:unit` or `npm test` — too slow (642 files, >4 min). Confirm: {{goal}} is achieved. Start the report with the verdict line `PASS: <what verified> — <evidence>` or `FAIL: <what failed> — <reason>` (the green-contract consumer reads it; no verdict line = UNVERIFIED).",
 			maxTasks: 1,
 			dependsOn: ["review"],
 			verificationCommand: "npm run test:critical && npx tsc --noEmit",
@@ -189,7 +189,7 @@ registerPlanTemplate({
 			name: "verify",
 			role: "verifier",
 			taskTemplate:
-				"Verify the complete implementation of: {{goal}}. Run FAST checks (`npm run test:critical && npx tsc --noEmit`, completes in <2 min). Do NOT run `npm run test:unit` or `npm test` — too slow for in-loop verification (642 files, >4 min). Validate all acceptance criteria.",
+				"Verify the complete implementation of: {{goal}}. Run FAST checks (`npm run test:critical && npx tsc --noEmit`, completes in <2 min). Do NOT run `npm run test:unit` or `npm test` — too slow for in-loop verification (642 files, >4 min). Validate all acceptance criteria. Start the report with the verdict line `PASS: <what verified> — <evidence>` or `FAIL: <what failed> — <reason>` (the green-contract consumer reads it; no verdict line = UNVERIFIED).",
 			maxTasks: 1,
 			dependsOn: ["review"],
 			verificationCommand: "npm run test:critical && npx tsc --noEmit",

package/src/runtime/task-runner/post-execution.ts

@@ -127,11 +127,17 @@ export type VerifierVerdict = "pass" | "fail" | "none";
  * (`live-session-runtime.ts` buildOutputContract → "PASS: <what verified> —
  * <evidence>" / "FAIL: <what failed> — <reason>"). Tolerates the markdown
  * decorations real workers emit (`**PASS:**`, `- FAIL:`, `> PASS:`,
- * `` `PASS:` ``, `# PASS:`), but only at line start — a bare mention of the
- * word inside prose does NOT count, and the colon is REQUIRED (the contract
- * has one; a colon-less "PASS" is prose, not a verdict).
+ * `` `PASS:` ``, `# PASS:`) AND the bare-prose verdicts the child-process
+ * workflow prompts used to ask for ("Give PASS or FAIL with evidence" →
+ * `PASS`, `**PASS**`, `PASS — evidence`, `PASS. Evidence: …`): a wave-6 T7
+ * live probe caught a FAILED-gate run blocking as `green_unsatisfied /
+ * observed=none` because the verifier followed its prose prompt and wrote a
+ * bare verdict. A token only counts at LINE START and only when followed by
+ * punctuation/EOL/`(` — a bare mention inside prose ("the verifier said
+ * PASS: that") still does NOT count, and a word continuation ("PASSED all
+ * checks") does not either.
  */
-const VERIFIER_VERDICT_LINE = /^[\s*_>#`=-]*(PASS|FAIL)\s*:/i;
+const VERIFIER_VERDICT_LINE = /^[\s*_>#`=-]*(PASS|FAIL)[\s*_`]*(?=[:.\u2014\u2013-]|$|\()/i;
 
 /**
  * M3-2 (R47-3/V4, fix-wave 6): read the verifier verdict out of the

package/workflows/default.workflow.md

@@ -36,6 +36,6 @@ Rules:
 - **MAXIMIZE PARALLELISM**: Put independent tasks in the SAME phase so they run concurrently. Never create sequential phases when tasks are independent.
 - Task titles name the concrete piece of work — they become the task list the user watches.
 - Choose the smallest effective number of tasks per phase; a simple goal may have 1-2 phases with 1-2 tasks.
-- End implementation plans with a verification phase (`verifier` role): run FAST targeted checks only (e.g. the affected test files), never the full suite.
+- End implementation plans with a verification phase (`verifier` role): run FAST targeted checks only (e.g. the affected test files), never the full suite. The verifier's report MUST start with the verdict line — `PASS: <what verified> — <evidence>` or `FAIL: <what failed> — <reason>` (the green-contract consumer reads it; a report with no verdict line is treated as UNVERIFIED).
 - Tasks within the same phase run in parallel; phases run sequentially.
 - Do not include more than 12 tasks per phase; split or summarize oversized plans instead.

package/workflows/fast-fix.workflow.md

@@ -21,4 +21,4 @@ dependsOn: execute
 verify: true
 
 Verify the fix with available evidence.
-Run FAST checks ONCE and cache output WITH provenance to `.crew/cache/` (per `agents/verifier.md`: unique per-attempt filename, `set -o pipefail` so the recorded exit code is the command's — not `tee`'s — and a sibling `.meta` recording command, git revision, tree fingerprint): `npm run test:critical && npx tsc --noEmit` (completes in <60s; if this repo lacks these scripts, use its fastest documented check instead). Do NOT run `npm run test:unit` or `npm test` — too slow (642 files, >4 min). Cross-reference cached output with the fix — reuse a cached log only when its provenance matches. Do NOT re-run tests. Give PASS or FAIL with specific test evidence.
+Run FAST checks ONCE and cache output WITH provenance to `.crew/cache/` (per `agents/verifier.md`: unique per-attempt filename, `set -o pipefail` so the recorded exit code is the command's — not `tee`'s — and a sibling `.meta` recording command, git revision, tree fingerprint): `npm run test:critical && npx tsc --noEmit` (completes in <60s; if this repo lacks these scripts, use its fastest documented check instead). Do NOT run `npm run test:unit` or `npm test` — too slow (642 files, >4 min). Cross-reference cached output with the fix — reuse a cached log only when its provenance matches. Do NOT re-run tests. Start your report with the verdict line — `PASS: <what verified> — <evidence>` or `FAIL: <what failed> — <reason>` (exact single-line format; the leader's green-contract consumer reads this line — a report with no verdict line is treated as UNVERIFIED). Then give details.

package/workflows/plan-execute.workflow.md

@@ -27,4 +27,4 @@ dependsOn: execute
 verify: true
 
 Verify completion for: {goal}
-Run FAST checks ONCE and cache output WITH provenance to `.crew/cache/` (per `agents/verifier.md`: unique per-attempt filename, `set -o pipefail` so the recorded exit code is the command's — not `tee`'s — and a sibling `.meta` recording command, git revision, tree fingerprint): `npm run test:critical && npx tsc --noEmit` (completes in <60s; if this repo lacks these scripts, use its fastest documented check instead). Do NOT run `npm run test:unit` or `npm test` — too slow (642 files, >4 min). Cross-reference cached output with the changes — reuse a cached log only when its provenance matches. Do NOT re-run tests. Give PASS or FAIL with specific test evidence.
+Run FAST checks ONCE and cache output WITH provenance to `.crew/cache/` (per `agents/verifier.md`: unique per-attempt filename, `set -o pipefail` so the recorded exit code is the command's — not `tee`'s — and a sibling `.meta` recording command, git revision, tree fingerprint): `npm run test:critical && npx tsc --noEmit` (completes in <60s; if this repo lacks these scripts, use its fastest documented check instead). Do NOT run `npm run test:unit` or `npm test` — too slow (642 files, >4 min). Cross-reference cached output with the changes — reuse a cached log only when its provenance matches. Do NOT re-run tests. Start your report with the verdict line — `PASS: <what verified> — <evidence>` or `FAIL: <what failed> — <reason>` (exact single-line format; the leader's green-contract consumer reads this line — a report with no verdict line is treated as UNVERIFIED). Then give details.

package/workflows/review.workflow.md

@@ -28,4 +28,4 @@ role: verifier
 dependsOn: code-review, security-review
 verify: true
 
-Run FAST checks ONCE and cache output WITH provenance to `.crew/cache/` (per `agents/verifier.md`: unique per-attempt filename, `set -o pipefail` so the recorded exit code is the command's — not `tee`'s — and a sibling `.meta` recording command, git revision, tree fingerprint): `npm run test:critical && npx tsc --noEmit` (completes in <60s; if this repo lacks these scripts, use its fastest documented check instead). Do NOT run `npm run test:unit` or `npm test` — too slow (642 files, >4 min). Cross-reference cached output with reviewer and security-reviewer findings — reuse a cached log only when its provenance matches. Confirm each finding against real test output. Give PASS if findings match evidence, FAIL if critical findings are false positives or tests reveal new issues.
+Run FAST checks ONCE and cache output WITH provenance to `.crew/cache/` (per `agents/verifier.md`: unique per-attempt filename, `set -o pipefail` so the recorded exit code is the command's — not `tee`'s — and a sibling `.meta` recording command, git revision, tree fingerprint): `npm run test:critical && npx tsc --noEmit` (completes in <60s; if this repo lacks these scripts, use its fastest documented check instead). Do NOT run `npm run test:unit` or `npm test` — too slow (642 files, >4 min). Cross-reference cached output with reviewer and security-reviewer findings — reuse a cached log only when its provenance matches. Confirm each finding against real test output. Start your report with the verdict line — `PASS: <what verified> — <evidence>` or `FAIL: <what failed> — <reason>` (exact single-line format; the leader's green-contract consumer reads this line — a report with no verdict line is treated as UNVERIFIED). Then give details.

package/workflows/strict-fast-fix.workflow.md

@@ -23,4 +23,4 @@ dependsOn: execute
 verify: true
 
 Verify the fix with available evidence, AND cross-reference the SPEC-EVIDENCE footer from explore+execute against the spec's requirements (machine-check the citations vs the spec id set).
-Run FAST checks ONCE and cache output WITH provenance to `.crew/cache/` (per `agents/verifier.md`: unique per-attempt filename, `set -o pipefail` so the recorded exit code is the command's — not `tee`'s — and a sibling `.meta` recording command, git revision, tree fingerprint): `npm run test:critical && npx tsc --noEmit` (completes in <60s; if this repo lacks these scripts, use its fastest documented check instead). Do NOT run `npm run test:unit` or `npm test` — too slow (642 files, >4 min). Cross-reference cached output with the fix — reuse a cached log only when its provenance matches. Do NOT re-run tests. Give PASS or FAIL with specific test evidence + spec gate verdict.
+Run FAST checks ONCE and cache output WITH provenance to `.crew/cache/` (per `agents/verifier.md`: unique per-attempt filename, `set -o pipefail` so the recorded exit code is the command's — not `tee`'s — and a sibling `.meta` recording command, git revision, tree fingerprint): `npm run test:critical && npx tsc --noEmit` (completes in <60s; if this repo lacks these scripts, use its fastest documented check instead). Do NOT run `npm run test:unit` or `npm test` — too slow (642 files, >4 min). Cross-reference cached output with the fix — reuse a cached log only when its provenance matches. Do NOT re-run tests. Start your report with the verdict line — `PASS: <what verified> — <evidence>` or `FAIL: <what failed> — <reason>` (exact single-line format; the leader's green-contract consumer reads this line — a report with no verdict line is treated as UNVERIFIED). Then give details. Include the spec-gate verdict too.