@llipe.com/dev-tasks 0.16.1 → 0.16.2

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
@@ -109,15 +109,17 @@ If the user provides a feature description or asks to create a PRD/spec/stories
109
109
  - Run `git rev-parse --abbrev-ref HEAD` to determine the current branch.
110
110
  - If HEAD is the default branch (`main`) or does not match `issue/*` or `story/*` pattern, you **MUST** create a new branch before proceeding, then open the Draft PR immediately after the first commit on that branch. You **MUST NOT** write implementation code, create files, or make commits while on the default branch.
111
111
  - If HEAD is already a valid feature branch (matching `issue/*` or `story/*`), confirm the Draft PR exists and proceed. If the branch has no commits yet, no PR can exist — open it immediately after the first commit.
112
- 5. Execute one sub-task at a time in checklist order. For each behavioral sub-task, follow **test-first**: write/update tests first, verify they fail for the right reason, then implement to make them pass.
113
- 6. After each completed sub-task: mark `[x]` locally and in GitHub, pause for approval if step-gated.
114
- 7. When all sub-tasks are complete:
112
+ 5. **memo intent (when memo is available):** Before writing any implementation code, run the memo-cli Availability Check and write the intent entry (see **memo-cli Integration → Intent Entry**). Record `memo_intent` as `written(<id>)`, `failed(<reason>)`, or `skipped(<reason>)`.
113
+ 6. Execute one sub-task at a time in checklist order. For each behavioral sub-task, follow **test-first**: write/update tests first, verify they fail for the right reason, then implement to make them pass.
114
+ 7. After each completed sub-task: mark `[x]` locally and in GitHub, pause for approval if step-gated.
115
+ 8. When all sub-tasks are complete:
115
116
  - Verify all acceptance criteria.
116
117
 
117
118
  - Run mandatory quality gates and record results (`test`, `lint`, `format:check`, `typecheck`, `audit`; `validate` if available).
118
119
  - For migration-bearing changes, confirm migration artifact/rollback notes and execute apply only after explicit user confirmation.
119
120
  - Record `coverage_gate` and `verifier_audit` per rules 22 and 18 above: `SKIPPED(no-delegation)` / `not-run(no-delegation)` by default, or the real result obtained under the Main-Thread Mode Addendum.
120
121
  - Record `docs_drift_status` per rule 10 above.
122
+ - Write the memo outcome entry (see **memo-cli Integration → Outcome Entry**) and record `memo_outcome` as `written(<id>)`, `failed(<reason>)`, or `skipped(<reason>)`.
121
123
  - Convert PR from Draft to Ready for Review once the applicable gates above are satisfied (by the caller, in the default subagent-delegation context, or by this agent itself under the addendum).
122
124
 
123
125
  ---
@@ -184,11 +186,9 @@ Everything above describes this agent's default operating context: a `Task`-dele
184
186
 
185
187
  ### Bank Declaration
186
188
 
187
- This agent's bank id is `developer-memory`, stable for the agent's life (PRD §15). Export it before any memo command in the session:
189
+ This agent's bank id is `developer-memory`, stable for the agent's life (PRD §15). Every memo command below names it **literally** with `--bank developer-memory`.
188
190
 
189
- ```bash
190
- export MEMO_BANK=developer-memory
191
- ```
191
+ You **MUST NOT** rely on `export MEMO_BANK=…` from an earlier command, and **MUST NOT** write `--bank $MEMO_BANK`. Agent runtimes run each shell command in a fresh shell, so an exported variable is gone by the next call; an empty `$MEMO_BANK` is dropped by the shell and the flag parser takes the next token as the bank id (`--bank --rationale …` → bank `--rationale`), which memo-cli rejects with `VALIDATION_FAILED`. This is the defect that left consumer banks empty (issue #253).
192
192
 
193
193
  ### Availability Check
194
194
 
@@ -198,18 +198,19 @@ At the start of every execution session, check if memo-cli is configured:
198
198
  which memo && memo setup validate
199
199
  ```
200
200
 
201
- - If `memo` is not found, skip all memo operations silently.
202
- - If `memo` is found but validation fails, ask: "memo-cli is installed but not configured for this repository. Run `memo setup init --repo <repo> --org <org> --domain <domain>` to configure it."
201
+ - If `memo` is not found **and** `memo.config.json` does not exist, skip all memo operations and record `memo_intent: skipped(memo-not-installed)` / `memo_outcome: skipped(memo-not-installed)`.
202
+ - If `memo` is not found **but** `memo.config.json` exists, the repository expects memo: warn the user ("memo-cli is configured for this repository (`memo.config.json`) but the `memo` binary is not on PATH in this session — no memo entries will be written.") and record `skipped(memo-unavailable)` for both fields. Never skip silently in this case.
203
+ - If `memo` is found but validation fails, ask: "memo-cli is installed but not configured for this repository. Run `memo setup init --repo <repo> --org <org> --domain <domain>` to configure it." If the run continues without it, record `skipped(memo-not-configured)`.
203
204
 
204
205
  ### Session Start — Restore Context
205
206
 
206
207
  When memo is available and `memo --version` is `1.3.0` or later, restore context with a single call:
207
208
 
208
209
  ```bash
209
- memo recall "<story or issue description>" --bank $MEMO_BANK --json
210
+ memo recall "<story or issue description>" --bank developer-memory --json
210
211
  ```
211
212
 
212
- `recall` returns one bundle — SELF, POLICIES, SHARED, MINE, LAST SESSION, CONFLICTS — in a single `<4s` call (SELF/MINE/LAST SESSION are omitted when `$MEMO_BANK` is unset or resolves to `kb`). If `memo recall` exits `2` (embeddings failure), surface the warning on stderr and continue the session without recalled context rather than blocking.
213
+ `recall` returns one bundle — SELF, POLICIES, SHARED, MINE, LAST SESSION, CONFLICTS — in a single `<4s` call. If `memo recall` exits `2` (embeddings failure), surface the warning on stderr and continue the session without recalled context rather than blocking.
213
214
 
214
215
  **Fallback — `memo --version` is below `1.3.0`:** run the four-command sequence instead:
215
216
 
@@ -240,45 +241,55 @@ Choose the most specific entry type:
240
241
 
241
242
  ### Intent Entry — Before Starting a Story
242
243
 
243
- Write an intent entry **before beginning implementation** of any story or issue, as an episodic entry scoped to this session in this agent's bank:
244
+ Write an intent entry **before beginning implementation** of any story or issue (Execution Flow step 5), as an episodic entry scoped to this session in this agent's bank:
244
245
 
245
246
  ```bash
246
247
  memo write \
247
248
  --kind episodic \
248
249
  --session ISSUE-<n> \
249
- --bank $MEMO_BANK \
250
+ --bank developer-memory \
250
251
  --rationale "Context: Starting ISSUE-<##> because <trigger/need>. Decision: implement via <approach>, preserving <constraints/non-goals>, with expected files <key files>. Impact: affects <user/module/contract impact> and introduces risks <if any>." \
251
252
  --tags "<domain>,issue-<number>,intent,<impact-tag>[,<boundary-tag>]" \
252
- --entry-type decision \
253
- --source agent \
254
- --story "ISSUE-<number>" \
255
- --on-duplicate consolidate \
256
- --json
253
+ --entry-type decision \
254
+ --source agent \
255
+ --story "ISSUE-<number>" \
256
+ --on-duplicate consolidate \
257
+ --json
257
258
  ```
258
259
 
259
260
  ### Outcome Entry — After Completing a Story
260
261
 
261
- Write an outcome entry as part of the **Completion Gate**, after all tests pass and before converting the PR to Ready for Review, as an episodic entry scoped to this session in this agent's bank:
262
+ Write an outcome entry as part of the **Completion Gate** (Execution Flow step 8), after all tests pass and before converting the PR to Ready for Review, as an episodic entry scoped to this session in this agent's bank. Record what was learned, not only what shipped — a rejected approach, a surprise in the codebase, or a constraint discovered mid-story is the most valuable content of this entry:
262
263
 
263
264
  ```bash
264
265
  memo write \
265
266
  --kind episodic \
266
267
  --session ISSUE-<n> \
267
- --bank $MEMO_BANK \
268
- --rationale "Context: Completed ISSUE-<##> after implementing <scope>. Delivery: shipped <behavior>, deviations <none|details>, AC <x/y> verified. Impact: quality gates test=<pass|fail>; lint=<pass|fail>; format:check=<pass|fail>; typecheck=<pass|fail>; audit=<pass|fail>; docs=<clean|drift-fixed>; migration=<none|details>." \
269
- --tags "<domain>,issue-<number>,outcome,gates-pass[,<impact-tag>][,<boundary-tag>]" \
270
- --entry-type decision \
271
- --source agent \
272
- --commit "$(git rev-parse HEAD)" \
273
- --story "ISSUE-<number>" \
274
- --files "<comma-separated key files modified>" \
275
- --on-duplicate consolidate \
276
- --json
268
+ --bank developer-memory \
269
+ --rationale "Context: Completed ISSUE-<##> after implementing <scope>. Delivery: shipped <behavior>, deviations <none|details>, AC <x/y> verified. Learned: <rejected approaches, surprises, constraints discovered|none>. Impact: quality gates test=<pass|fail>; lint=<pass|fail>; format:check=<pass|fail>; typecheck=<pass|fail>; audit=<pass|fail>; docs=<clean|drift-fixed>; migration=<none|details>." \
270
+ --tags "<domain>,issue-<number>,outcome,gates-pass[,<impact-or-boundary-tag>]" \
271
+ --entry-type decision \
272
+ --source agent \
273
+ --commit "$(git rev-parse HEAD)" \
274
+ --story "ISSUE-<number>" \
275
+ --files "<comma-separated key files modified>" \
276
+ --on-duplicate consolidate \
277
+ --json
277
278
  ```
278
279
 
280
+ memo-cli accepts 2–5 tags per entry and rejects more with `VALIDATION_FAILED`; keep both templates within that limit.
281
+
279
282
  Session sequence numbers (`--seq`) auto-increment per `--session`; do not pass `--seq` explicitly unless replaying a specific position.
280
283
 
281
- ADR and durable-decision entries are **not** written by `developer` — they stay `--kind semantic` in `kb`, authored by `technical-writer` with `--provenance <csv>` (citing the episodic ids that informed the decision) or `--manual`, per PRD K3.
284
+ ### Write Results Are Recorded, Never Swallowed
285
+
286
+ Every `memo write` above returns JSON with the new entry's `id` on success, and a non-zero exit with an error on failure.
287
+
288
+ - On success, record the id: `memo_intent: written(<id>)` / `memo_outcome: written(<id>)`.
289
+ - On failure, read the error, fix the command if the error names a flag or value you supplied, and retry **once**. If it still fails, record `failed(<error code or first line of the error>)` and state it in the closeout summary. A failed memo write does not block the story, but it **MUST NOT** be dropped silently.
290
+ - Both fields are **mandatory** in the closeout payload. Omitting either one marks the story incomplete.
291
+
292
+ ADR and durable-decision entries are **not** written by `developer` — they stay `--kind semantic` in `kb`, authored by `technical-writer` and `product-engineer`, per PRD K3.
282
293
 
283
294
  ### Session Close
284
295
 
@@ -299,7 +310,7 @@ Before marking a Story/Issue done:
299
310
  7. `technical-writer` agent **MUST** have run and produced both a delta report and a drift/stale-doc validation result.
300
311
  8. `/docs` **MUST** be updated to current state.
301
312
  9. `/workstream` **SHOULD** be cleaned (active artifacts retained, obsolete artifacts archived/removed).
302
- 10. If memo-cli is available, an episodic outcome entry **MUST** be written to `$MEMO_BANK` (`memo write --kind episodic --session ISSUE-<n> --bank $MEMO_BANK …`) before PR conversion.
313
+ 10. The memo intent and outcome writes **MUST** have been attempted when memo-cli is available (`memo write --kind episodic --session ISSUE-<n> --bank developer-memory …`, before PR conversion), and `memo_intent` / `memo_outcome` **MUST** be recorded as `written(<id>)`, `failed(<reason>)`, or `skipped(<reason>)`. A `failed` or `skipped` value does not block completion; an omitted field does.
303
314
  11. PR **MUST** be ready, approved, and merged.
304
315
  12. You **MUST NOT** close the GitHub Issue until all conditions above are met.
305
316
  13. For multi-story implementations, you **MUST** run a checklist cross-check between GitHub Issue tasks and `/workstream/tasks-*.md` and report any mismatch resolution.
@@ -360,6 +371,8 @@ workstream_files:
360
371
  - typecheck: PASS | FAIL | NOT RUN
361
372
  - audit: PASS | FAIL | NOT RUN
362
373
  coverage_gate: PASS | FAIL | SKIPPED(<reason>)
374
+ memo_intent: written(<id>) | failed(<reason>) | skipped(<reason>)
375
+ memo_outcome: written(<id>) | failed(<reason>) | skipped(<reason>)
363
376
  checklist_sync: synced | mismatch-fixed | blocked
364
377
  verifier_audit: run | not-run(no-delegation) | blocked
365
378
  fidelity_verdict: High | Medium | Low | none
@@ -373,6 +386,7 @@ Rules for this payload:
373
386
 
374
387
  - The markers `BEGIN CLOSEOUT PAYLOAD` and `END CLOSEOUT PAYLOAD` **MUST** appear exactly as written.
375
388
  - Every field is required. Use `none`, `NOT RUN`, or `blocked` when a value does not exist.
389
+ - `memo_intent` and `memo_outcome` are always present. `skipped(<reason>)` (memo not installed, not configured, or unavailable in this session) and `failed(<reason>)` (the write was attempted and errored after one retry) are valid values that do not block the merge; `planner` treats an absent field as an incomplete story.
376
390
  - `planner` may treat the story as incomplete if either marker or any required field is missing.
377
391
  - `verifier_audit: not-run(no-delegation)` and `coverage_gate: SKIPPED(no-delegation)` are the honest, expected values in this agent's default subagent-delegation context (see Operating Context above) — `planner` treats these as a signal to run its own direct invocation, not as a red flag or an incomplete story.
378
392
 
@@ -325,13 +325,9 @@ Ensure:
325
325
 
326
326
  ### Bank Declaration
327
327
 
328
- This agent's bank id is `technical-writer-memory`, stable for the agent's life (PRD §15). Export it before any memo command in the session:
328
+ This agent's bank id is `technical-writer-memory`, stable for the agent's life (PRD §15). Name banks **literally** in every memo command (`--bank kb`, `--bank technical-writer-memory`); never rely on an `export MEMO_BANK=…` from an earlier command and never write `--bank $MEMO_BANK` — each shell command runs in a fresh shell, the variable is empty, and memo-cli rejects the resulting bank id (issue #253).
329
329
 
330
- ```bash
331
- export MEMO_BANK=technical-writer-memory
332
- ```
333
-
334
- ADR and decision entries this agent writes stay `--kind semantic` in `kb` (never in `$MEMO_BANK`) — see **Write Template** below.
330
+ ADR and decision entries this agent writes stay `--kind semantic` in `kb` (never in its private bank) — see **Write Template** below.
335
331
 
336
332
  ### Availability Check
337
333
 
@@ -341,7 +337,8 @@ At the start of every run, check if memo-cli is configured:
341
337
  which memo && memo setup validate
342
338
  ```
343
339
 
344
- - If `memo` is not found, skip all memo operations silently.
340
+ - If `memo` is not found and `memo.config.json` does not exist, skip all memo operations.
341
+ - If `memo` is not found but `memo.config.json` exists, warn the user that the repository expects memo-cli but the binary is not on PATH in this session, and report `memo: skipped(memo-unavailable)` in the run summary. Never skip silently in this case.
345
342
  - If `memo` is found but validation fails (missing `memo.config.json` or env vars), **STOP** and ask: "memo-cli is installed but not configured for this repository. Run `memo setup init --repo <repo> --org <org> --domain <domain>` to configure it, then re-run."
346
343
 
347
344
  ### Read Before Write
@@ -382,7 +379,7 @@ memo write \
382
379
  --tags "<domain-tag>,<entry-nature-tag>,<story-ref-if-known>" \
383
380
  --entry-type <decision|structure|integration_point> \
384
381
  --source agent \
385
- --provenance "<csv of episodic entry ids this decision consolidates, if any>" \
382
+ --provenance "<csv of episodic entry ids this decision consolidates>" \
386
383
  --commit "$(git rev-parse HEAD)" \
387
384
  --story "<issue-or-story-id-if-known>" \
388
385
  --files "<path/to/doc.md>,<path/to/adr.md>" \
@@ -390,7 +387,12 @@ memo write \
390
387
  --json
391
388
  ```
392
389
 
393
- ADR and durable-decision entries are always `--kind semantic` in `kb`, never in `$MEMO_BANK` — this agent is the one long-lived agent that writes to the shared `kb`, not a private bank. Per PRD K3, a `semantic` entry with `--source agent` **MUST** carry `--provenance <csv>` (the episodic entry ids — typically `developer`'s intent/outcome entries — that informed the decision) unless `--manual` is passed instead (human-authored, no episodic origin).
390
+ ADR and durable-decision entries are always `--kind semantic` in `kb`, never in a private bank. Per PRD K3, memo-cli **rejects** a `semantic` entry with `--source agent` and no provenance (`VALIDATION_FAILED`), so choose exactly one form:
391
+
392
+ - **Provenance known:** find the episodic ids that informed the decision — typically `developer`'s intent/outcome entries for the story — with `memo list --bank developer-memory --session ISSUE-<n> --json`, and pass them as `--source agent --provenance "<id1>,<id2>"`.
393
+ - **No episodic ids found:** drop both `--source agent` and `--provenance`, and pass `--manual` instead. Never pass an empty `--provenance ""`.
394
+
395
+ **Write results are recorded, never swallowed.** On success, the JSON response carries the entry `id`; list every written id in the run summary. On failure, fix the command if the error names a flag or value you supplied and retry once; if it still fails, report `memo: failed(<error>)` in the run summary. A failed write does not block the documentation run, but it **MUST NOT** be dropped silently.
394
396
 
395
397
  **Rationale quality rule:** Answer three questions in one coherent paragraph: (1) what changed and why it was needed, (2) what was decided or documented, (3) what it affects downstream. Never use bullet points inside `--rationale`.
396
398
 
@@ -247,6 +247,21 @@ new category. Never hold a PR on it.
247
247
  | Test-harness setup | `qa-engineer` |
248
248
  | Tooling setup | `housekeeping` |
249
249
 
250
+ ## Roadmap-Mismatch Finding
251
+
252
+ `docs/roadmap.md` is the one-page index of every PRD's phases or waves and their status. It goes stale silently unless something compares it with what actually merged, so that comparison lives here as a finding.
253
+
254
+ **Trigger** (PRD-level rollup audits, and story audits whose story completes a phase or wave). Report the finding when any of these hold:
255
+
256
+ 1. The audited work completes a PRD phase or wave, and its roadmap row is not `Done` in the delivered tree.
257
+ 2. A roadmap row is `Done`, but the audited scope shows that phase's stories unmerged or its acceptance criteria unmet.
258
+ 3. The audited PRD defines a phase or wave with no roadmap row, or `docs/roadmap.md` is missing while `docs/requirements/` holds a PRD.
259
+ 4. A row's `Issues/PRs` or `What's missing` contradicts the audited issues and PRs.
260
+
261
+ **Not a finding:** rows for PRDs outside the audited scope; a row left `In progress` while some of its stories are still open.
262
+
263
+ **The finding is advisory.** Like every other drift item it is **non-blocking to PR readiness and to issue completion**, and it routes to `product-engineer`'s `activity-drift-reconciliation` flow, which corrects the row. Classify it with the existing impact/intent vocabulary. Never hold a PR on it.
264
+
250
265
  ## Failure Triage Workflow (Randomized Tests)
251
266
 
252
267
  When a randomized or fuzz test fails, follow this sequence:
@@ -60,13 +60,39 @@ If any required input is missing, ask one focused question with a default option
60
60
 
61
61
  ### Bank Declaration
62
62
 
63
- This agent's bank id is `planner-memory`, stable for the agent's life (PRD §15). Export it before any memo command in the session:
63
+ This agent's bank id is `planner-memory`, stable for the agent's life (PRD §15). Name it **literally** in every memo command (`--bank planner-memory`); never rely on an `export MEMO_BANK=…` from an earlier command and never write `--bank $MEMO_BANK` — each shell command runs in a fresh shell, the variable is empty, and memo-cli rejects the resulting bank id (issue #253).
64
+
65
+ ### Availability Check
66
+
67
+ At the start of every run:
64
68
 
65
69
  ```bash
66
- export MEMO_BANK=planner-memory
70
+ which memo && memo setup validate
67
71
  ```
68
72
 
69
- `planner` does not read or write memo entries itself — orchestration-scoped context comes from the task file/GitHub milestone, and each delegated `developer` subagent runs its own `memo recall`/`memo write` sequence scoped to `developer-memory`. This declaration exists for AC parity with the other long-lived agent definitions and for any future orchestration-level use of `--bank $MEMO_BANK`.
73
+ - If `memo` is not found and `memo.config.json` does not exist, skip all memo operations.
74
+ - If `memo` is not found but `memo.config.json` exists, warn the user that the repository expects memo-cli but the binary is not on PATH in this session, and record `memo_run_outcome: skipped(memo-unavailable)`. Never skip silently in this case.
75
+
76
+ ### Responsibilities
77
+
78
+ - **Per story — verify, don't write.** Each delegated `developer` subagent writes its own intent/outcome entries to `developer-memory`. Planner verifies them through the closeout payload's `memo_intent` / `memo_outcome` fields (merge gate 3). An absent field means the story is incomplete; `skipped(<reason>)` or `failed(<reason>)` is reported in the story status table and does not block the merge.
79
+ - **Per run — write one outcome entry.** Before opening the consolidated PR (Phase 5 step 6), write one episodic entry summarizing the run:
80
+
81
+ ```bash
82
+ memo write \
83
+ --kind episodic \
84
+ --session <plan-id> \
85
+ --bank planner-memory \
86
+ --rationale "Context: Ran <plan-id> (<PRD/milestone>, <roadmap phase/wave>). Delivery: <n> stories merged, <n> blocked, integration branch <branch>. Learned: <recurring failures, rework, dependency surprises, gates that caught real problems|none>. Memo coverage: <n>/<n> stories with written intent+outcome; <skipped/failed summary|none>." \
87
+ --tags "<domain>,<plan-id>,run-outcome,<roadmap-phase-tag>" \
88
+ --entry-type decision \
89
+ --source agent \
90
+ --commit "$(git rev-parse HEAD)" \
91
+ --on-duplicate consolidate \
92
+ --json
93
+ ```
94
+
95
+ Record the result as `memo_run_outcome: written(<id>) | failed(<reason>) | skipped(<reason>)` in the planner state file and the consolidated PR body. On failure, fix the command if the error names a flag or value you supplied and retry once; never drop a failure silently.
70
96
 
71
97
  ### Session Close
72
98
 
@@ -111,6 +137,10 @@ Expected return for each issue:
111
137
 
112
138
  Map milestone issues into the same internal story model used by Option A.
113
139
 
140
+ ### Roadmap Read
141
+
142
+ Read `docs/roadmap.md` (the roadmap overview — see **Roadmap Maintenance** below). Identify the PRD and the phase/wave row this run delivers, and state it when confirming the source. If the file is missing, or no row matches this run's scope, say so and ask `product-engineer` (or the user) to add the row before Phase 4; do not invent a row's scope yourself.
143
+
114
144
  Confirm source and story count before Phase 0.5.
115
145
 
116
146
  ---
@@ -339,6 +369,7 @@ Integration target branch: {{ integration_branch }}
339
369
  Test-first: YES — write/update tests before implementation for each behavioral sub-task
340
370
  Test plan: {{ test_plan_path | default: "none — derive tests from acceptance criteria" }}
341
371
  Decision log: {{ decision_log_path }}
372
+ Memo: if memo-cli is available, write the intent entry before coding and the outcome entry before PR conversion, with `--bank developer-memory` named literally; return `memo_intent` and `memo_outcome` in the closeout payload
342
373
 
343
374
  Implement only this story scope.
344
375
  Follow test-first design: for each behavioral sub-task, write tests first, verify they fail, then implement.
@@ -424,7 +455,7 @@ For each completed story PR:
424
455
 
425
456
  1. Verify PR base branch is the integration branch.
426
457
  2. Verify required checks are successful.
427
- 3. Verify delegated closeout payload reports `docs_drift_status` as `clean` or `drift-fixed`.
458
+ 3. Verify delegated closeout payload reports `docs_drift_status` as `clean` or `drift-fixed`, and carries both `memo_intent` and `memo_outcome` fields. An absent memo field marks the story incomplete (retry the closeout once, as for any missing payload field); a `skipped(<reason>)` or `failed(<reason>)` value is recorded in the story status table and does **NOT** block the merge.
428
459
  4. Verify delegated quality gates are all `PASS` (`test`, `lint`, `format:check`, `typecheck`, `audit`).
429
460
  5. **Invoke `qa-engineer` directly** — scoped to this story's diff/branch/PR — immediately after receiving the `developer` subagent's closeout payload. The `developer` subagent structurally lacks the `Task` tool (see `.claude/agents/developer.md`'s Operating Context) and may honestly self-report `coverage_gate: SKIPPED(no-delegation)`; that is expected in this path, not a red flag. Record the result of _planner's own_ `qa-engineer` invocation and verify it is `PASS`, `FAIL`, or `SKIPPED(<reason>)` carrying a non-empty reason. An omitted result means the QA gate was never reached — treat the story as incomplete. A `FAIL` or `SKIPPED` value does **NOT** block the merge; only omission does.
430
461
  6. **Invoke `verifier` in Audit Mode directly** — scoped to this story's diff/branch/PR — immediately after step 5, and post its human-readable summary to the story PR via `github-ops` comment conventions. A bare developer-reported `verifier_audit: run` string is one signal but is **NOT sufficient evidence on its own** — `verifier_audit: not-run(no-delegation)` is a valid, expected self-report from the `developer` subagent in this path (it structurally cannot invoke `verifier`), not a red flag. This merge gate checks the result of _planner's own_ `verifier` invocation, not the self-reported field. The audit's drift findings (`fidelity_verdict`/`highest_drift_impact`/`drift_findings`) **MUST NOT** block the merge.
@@ -461,14 +492,18 @@ After all stories are merged into integration:
461
492
  2. Invoke `qa-engineer` at **PRD scope** (all packages affected by the merged stories, all layers) for a coverage and gap rollup. Report a PRD-level `coverage_gate` that aggregates per-story gates. Scope to packages affected by the merged stories, not the entire workspace.
462
493
  3. Invoke `verifier` in `audit` mode for a **PRD-level rollup audit** against the full integrated scope (all merged stories on the integration branch, cross-checked against the source PRD/spec/milestone). This is mandatory and non-skippable — it runs whether or not any story-level drift was previously reported. Post the resulting human-readable summary to the plan/milestone issue via `github-ops` comment conventions. Rollup drift findings **MUST NOT** block the consolidated PR handoff; Unintended/Intended drift routes to `product-engineer`'s `activity-drift-reconciliation` flow.
463
494
  4. Invoke `technical-writer` for a planner-level drift/stale-doc validation pass against integrated changes; unresolved drift **MUST** block handoff.
464
- 5. Open one consolidated PR from integration branch to `main`.
465
- 6. **Do NOT merge.** Notify the user that the consolidated PR is ready for their review.
466
- 7. Wait for the user to approve and merge the PR into `main`.
467
- 8. **Post-integration deploy handoff (conditional).** When the merged scope includes deployable changes and the repository declares environments in `infra/environments.yaml`, planner **MUST** hand the deploy off to `infra-engineer` rather than running any platform write or deploy command itself. This handoff is conditional — it applies only when infrastructure/deploy scope is present; a docs- or test-only integration triggers no deploy. Planner names `infra-engineer` as the owner of the post-integration deploy (dev on merge to `main`, production behind the protected-environment reviewer on the release tag via the `deploy-ops` workflow templates) and never invokes `deploy.sh`, `release.sh`, `flyctl`, `aws`, or `supabase` directly.
468
- 9. Before final handoff, **MUST** ensure the local working branch is the integration branch used for this run:
469
- - Preferred: run `git checkout integration/<plan-id>-<short-description>`.
470
- - Alternative (if checkout is not possible in the current runtime): explicitly verify and report current branch, and provide the exact checkout command the user can run.
471
- 10. Final user response **MUST** include a `PR Directives (User Action Required)` section with:
495
+ 5. **Roadmap update (mandatory).** Update the run's row(s) in `docs/roadmap.md` on the integration branch, so the change ships in the consolidated PR itself: set `Status` (`Done` when every story in the phase/wave merged, otherwise `In progress`), fill `Issues/PRs` with the story issues and the consolidated PR, rewrite `What's missing` to the remaining scope (`—` when none), and set the `Last updated` date. A consolidated PR without this roadmap change is incomplete. See **Roadmap Maintenance** below.
496
+ 6. **memo run outcome.** Write the per-run outcome entry described in **memo-cli Integration → Responsibilities** and record `memo_run_outcome`.
497
+ 7. Open one consolidated PR from integration branch to `main`.
498
+ 8. **Do NOT merge.** Notify the user that the consolidated PR is ready for their review.
499
+ 9. Wait for the user to approve and merge the PR into `main`.
500
+ 10. **Post-integration deploy handoff (conditional).** When the merged scope includes deployable changes and the repository declares environments in `infra/environments.yaml`, planner **MUST** hand the deploy off to `infra-engineer` rather than running any platform write or deploy command itself. This handoff is conditional — it applies only when infrastructure/deploy scope is present; a docs- or test-only integration triggers no deploy. Planner names `infra-engineer` as the owner of the post-integration deploy (dev on merge to `main`, production behind the protected-environment reviewer on the release tag via the `deploy-ops` workflow templates) and never invokes `deploy.sh`, `release.sh`, `flyctl`, `aws`, or `supabase` directly.
501
+ 11. Before final handoff, **MUST** ensure the local working branch is the integration branch used for this run:
502
+
503
+ - Preferred: run `git checkout integration/<plan-id>-<short-description>`.
504
+ - Alternative (if checkout is not possible in the current runtime): explicitly verify and report current branch, and provide the exact checkout command the user can run.
505
+
506
+ 12. Final user response **MUST** include a `PR Directives (User Action Required)` section with:
472
507
 
473
508
  - consolidated PR URL
474
509
  - current CI/check status
@@ -482,6 +517,8 @@ Consolidated PR should include:
482
517
  - Per-story changed files and test results
483
518
  - Manual validation instructions per story
484
519
  - Integration test summary
520
+ - Roadmap row(s) updated in `docs/roadmap.md` (before → after status)
521
+ - Memo coverage: per-story `memo_intent` / `memo_outcome` values and `memo_run_outcome`
485
522
 
486
523
  PR title **MUST** follow Conventional Commits and PR body **MUST** follow `github-ops` conventions.
487
524
 
@@ -489,6 +526,17 @@ Planner **MUST NOT** merge the consolidated PR. Only the user may approve and me
489
526
 
490
527
  ---
491
528
 
529
+ ## Roadmap Maintenance
530
+
531
+ `docs/roadmap.md` is the one-page overview of every PRD's phases or waves: what is done, what is in progress, and what is missing. It is an index over the PRDs, GitHub issues, and merged PRs — when they disagree, they win and the roadmap is corrected.
532
+
533
+ - **Read** it in Phase 0 (see **Roadmap Read**).
534
+ - **Update** the run's row(s) in Phase 5 step 5, in the consolidated PR. Never leave the roadmap claiming `In progress` for a phase this run finished, or `Done` for one it did not.
535
+ - Change only the rows this run touched; rows for other PRDs belong to `product-engineer` and to the planner runs that deliver them.
536
+ - The `verifier` rollup audit (Phase 5 step 3) compares the roadmap against merged work and reports a mismatch as an advisory finding.
537
+
538
+ ---
539
+
492
540
  ## Error Handling
493
541
 
494
542
  | Situation | Action |
@@ -71,17 +71,38 @@ If any required input is missing, you **MUST** ask concise clarifying questions.
71
71
 
72
72
  ---
73
73
 
74
- ## memo-cli Integration (When Available)
74
+ ## Roadmap Maintenance
75
75
 
76
- ### Bank Declaration
76
+ `docs/roadmap.md` is the one-page overview of every PRD's phases or waves: what is done, what is in progress, and what is missing. `product-engineer` owns its rows and structure; `planner` updates the status of the rows a run delivers. It is an index over the PRDs, GitHub issues, and merged PRs — when they disagree, they win and the roadmap is corrected.
77
77
 
78
- This agent's bank id is `product-engineer-memory`, stable for the agent's life (PRD §15). Export it before any memo command in the session:
78
+ - **Read it at session start**, in every mode, before any activity, so a new PRD or issue is placed against what already exists and what is still missing.
79
+ - **Create it** in Init Mode, or the first time a PRD is written in a repository that has none, from the template shape below.
80
+ - **Add rows when a PRD is created or changed** (`activity-refine`): one row per delivery phase or wave the PRD defines, or a single `All` row when it defines none, each `Not started`. A PRD changelog that adds, removes, or re-scopes a phase updates its rows in the same change.
81
+ - **Fill `Issues/PRs` when stories are published** (`activity-publish-github`), and set the row to `In progress` once its first story is being implemented.
82
+ - **Correct it during drift reconciliation** (`activity-drift-reconciliation`) when a `verifier` roadmap-mismatch finding is confirmed.
83
+ - Never mark a row `Done` yourself on the strength of a plan; `Done` requires merged work, and `planner` sets it in the consolidated PR that delivers it.
79
84
 
80
- ```bash
81
- export MEMO_BANK=product-engineer-memory
85
+ Shape (one table, one row per PRD phase or wave, newest PRD first):
86
+
87
+ ```markdown
88
+ # Roadmap
89
+
90
+ Last updated: YYYY-MM-DD
91
+
92
+ | PRD | Phase/Wave | Scope (one line) | Status | Issues/PRs | What's missing |
93
+ | ---------------------------------------- | ---------- | ---------------- | ----------- | ----------- | ---------------- |
94
+ | [prd-<name>](requirements/prd-<name>.md) | 1 | <scope> | Not started | #12, PR #20 | <remaining or —> |
82
95
  ```
83
96
 
84
- `product-engineer` does not write episodic or semantic entries itself (see below), so this declaration exists for AC parity with the other long-lived agent definitions and for any future read-side use of `--bank $MEMO_BANK`.
97
+ `Status` is exactly one of `Not started`, `In progress`, `Done`.
98
+
99
+ ---
100
+
101
+ ## memo-cli Integration (When Available)
102
+
103
+ ### Bank Declaration
104
+
105
+ This agent's bank id is `product-engineer-memory`, stable for the agent's life (PRD §15). The decisions this agent records go to the shared `kb` (see **Decision Write-Back** below), not its private bank. Name the bank **literally** in every memo command (`--bank kb`); never rely on an `export MEMO_BANK=…` from an earlier command and never write `--bank $MEMO_BANK` — each shell command runs in a fresh shell, the variable is empty, and memo-cli rejects the resulting bank id (issue #253).
85
106
 
86
107
  ### Availability Check
87
108
 
@@ -91,7 +112,8 @@ At the start of every session, verify memo-cli is configured:
91
112
  which memo && memo setup validate
92
113
  ```
93
114
 
94
- - If `memo` is not found, skip all memo operations silently.
115
+ - If `memo` is not found and `memo.config.json` does not exist, skip all memo operations.
116
+ - If `memo` is not found but `memo.config.json` exists, warn the user that the repository expects memo-cli but the binary is not on PATH in this session, so no decisions will be recorded in memo. Never skip silently in this case.
95
117
  - If `memo` is found but validation fails, ask: "memo-cli is installed but not configured for this repository. Run `memo setup init --repo <repo> --org <org> --domain <domain>` to configure it."
96
118
 
97
119
  ### Session Start — Restore Context
@@ -126,7 +148,31 @@ Before making design choices in `activity-refine`, `activity-generate-spec`, or
126
148
  memo search "<specific topic, technology, or module>" --json
127
149
  ```
128
150
 
129
- **product-engineer does NOT write to memo.** All writes are delegated to `technical-writer` (ADRs and doc changes) and `developer` (intent/outcome entries).
151
+ ### Decision Write-Back — After Each Grilling Exit Gate
152
+
153
+ The decisions made with the user during refinement and specification are the most durable knowledge the workflow produces; they **MUST** reach memo, not only `workstream/decisions-<feature>.md`.
154
+
155
+ When `activity-grill`'s exit gate is satisfied (the WHAT phase in `activity-refine`, the HOW phase in `activity-generate-spec`, or Issue Mode refinement), write **one `kb` entry per user-answered decision row** recorded in that phase. Skip self-resolved rows (`Accepted rec.` is `n/a`): they restate knowledge that already exists. Do not batch several decisions into one entry.
156
+
157
+ ```bash
158
+ memo write \
159
+ --kind semantic \
160
+ --bank kb \
161
+ --manual \
162
+ --rationale "<feature>#D-NN (<WHAT|HOW>): <question in one sentence>. Decided: <answer>. Why: <reason given or the accepted recommendation's reasoning>. Rejected: <alternatives considered|none>." \
163
+ --tags "<domain>,<feature>,decision,<phase-lowercase>,<feature>-d-NN" \
164
+ --entry-type decision \
165
+ --story "<issue-or-feature-id-if-known>" \
166
+ --files "workstream/decisions-<feature>.md" \
167
+ --on-duplicate consolidate \
168
+ --json
169
+ ```
170
+
171
+ - `--manual` is correct here: the decision was made by the user in conversation, so it has no episodic provenance, and memo-cli rejects an agent-sourced semantic entry without one.
172
+ - A row that supersedes an earlier decision is written as a new entry naming the superseded ID in its rationale; `--on-duplicate consolidate` merges near-duplicates.
173
+ - **Write results are recorded, never swallowed.** Report the written ids (or `skipped(<reason>)`) in the phase's completion summary. On failure, fix the command if the error names a flag or value you supplied and retry once; then report `failed(<error>)`. A failed write never blocks the phase.
174
+
175
+ `technical-writer` still owns ADR and doc-change entries, and `developer` owns per-story intent/outcome entries.
130
176
 
131
177
  ### Session Close
132
178