mercury-agent 0.23.2 → 0.24.1

package/docs/plans/archive/2026-09-22-p0-bugs-0922/briefs/S2.md

@@ -0,0 +1,60 @@
+# Brief S2 — the system prompt advertises a URL-capable read the image lacks
+
+**Plan**: docs/plans/p0-bugs-0922/plan.md   **Session**: S2   **Round**: 1   **Kind**: bug
+**Item slug**: system-prompt-advertises-url-capable-read-the-image-lacks
+
+> The only file this session reads from the plan. Self-contained: it never points at a sibling's brief. The `**Item slug**:` line is the key the execution commands grep for to detect that a brief is active — keep it exactly in that shape.
+
+## Launch
+
+/f-ops-plan work p0-bugs-0922 S2
+
+## Run
+
+/f-bug-fix system-prompt-advertises-url-capable-read-the-image-lacks
+
+## Scope notes
+
+Cheapest correct fix: make the `read` line in `buildSystemPrompt`'s tools list say files only, and name the real way to fetch URLs (`curl` via bash, or pinchtab). **The bug's durable half is a test** that compares the snippet against the `description` of the `read` tool actually installed from the pinned `@earendil-works/pi-coding-agent`. Without it the snippet drifts again. Remember the whole-prompt-ownership gotcha (`--system-prompt`, never `--append-system-prompt`): edit the snippet text, not how the prompt is assembled. **Deploy needs: rebuild.** `container-entry.ts` is baked (`container/build.sh`, then restart; verify by grepping the image, never the repo). S3 in this plan also needs a rebuild, so one rebuild covers both.
+
+## Owned surfaces (predicted)
+
+- src/agent/container-entry.ts
+- container/agent-package.json
+
+## Do not touch
+
+- src/core/storage-cleanup.ts, src/storage/db.ts, src/core/reply-context.ts   (owned by S1 this plan)
+- resources/pi-extensions/subagent/index.ts   (owned by S3 this plan)
+- src/config.ts   (owned by S4 this plan)
+- Not ours (avishai tsabari / Avishai-Tsabari, last 14 d): `src/agent/container-entry.ts`, `src/config.ts`, `src/storage/db.ts`, `container/agent-package.json` — see plan.md "Not ours"; lane cleared only by the Q1/Q2 answers. Re-check with `/f-ops-plan next p0-bugs-0922`.
+
+## Edges
+
+- start-after: none   (satisfied — `next` launched you)
+- merge-after: none
+
+## Overlap verdicts
+
+none
+
+> The command's preflight overlap check treats a hit whose branch AND every shared file match a verdict line as answered (c), prints one `PLAN-PASSED(overlap): …` line, and carries it into the final report. Any other hit prompts as usual. The merge-time check is never waived.
+
+## Decisions answered
+
+- Q1: ours (owner, 2026-09-22) — `src/agent/container-entry.ts`, `src/config.ts`, `src/storage/db.ts` are this plan's lane.
+- Q2: ours (owner, 2026-09-22) — `container/agent-package.json` may be edited if the fix needs it.
+
+## Ledger rules
+
+- `docs/ROADMAP.md`: never edit. Your command writes `docs/pending-updates/system-prompt-advertises-url-capable-read-the-image-lacks.md` as usual.
+- `docs/pending-verification.md`: append only under `## Plan p0-bugs-0922 / S2`.
+- Any goal's `decisions.md`: never edit; put decisions in your log's "Siblings must know".
+
+## Environment
+
+- Read `CLAUDE.md` → Worktree Setup and Gotchas before opening the worktree. Nothing from them is copied here. `bun install` once in the fresh worktree; the check gate is `bun run check` (every link runs to completion — read every line).
+
+## Report
+
+Your command's final report is appended to `docs/plans/p0-bugs-0922/sessions/S2.md` by its report step (path resolved through `.claude/scripts/resolve-docs-root.sh docs/plans/probe.md`). `work` verifies the append.

package/docs/plans/archive/2026-09-22-p0-bugs-0922/briefs/S3.md

@@ -0,0 +1,58 @@
+# Brief S3 — subagent parallel mode reports API errors as completed
+
+**Plan**: docs/plans/p0-bugs-0922/plan.md   **Session**: S3   **Round**: 1   **Kind**: bug
+**Item slug**: subagent-parallel-mode-reports-api-errors-as-completed
+
+> The only file this session reads from the plan. Self-contained: it never points at a sibling's brief. The `**Item slug**:` line is the key the execution commands grep for to detect that a brief is active — keep it exactly in that shape.
+
+## Launch
+
+/f-ops-plan work p0-bugs-0922 S3
+
+## Run
+
+/f-bug-fix subagent-parallel-mode-reports-api-errors-as-completed
+
+## Scope notes
+
+One shared `isSubagentError(result)` predicate (`exitCode !== 0 || stopReason === "error" || stopReason === "aborted"`), used by chain, single and parallel, for both the behaviour (`successCount`, the `completed`/`failed` label) and the display (collapsed counters, ✓/✗ icon). **Re-read the line numbers.** The doc's references date from 2026-09-07 at `c7d0f30`. The checklist item "`resources/` inside typecheck + lint" may already be done (`gates-do-not-cover-resources-pi-extensions` was implemented 2026-09-07). Verify that it is and tick it; do not redo it. Test: `exitCode: 0, stopReason: "error"` counts as a failure in all three modes. Negative-test by reverting one call site and confirming that mode goes red. **Deploy needs: rebuild.** `resources/` is baked into the image. Verify inside the image, never the repo.
+
+## Owned surfaces (predicted)
+
+- resources/pi-extensions/subagent/index.ts
+
+## Do not touch
+
+- src/core/storage-cleanup.ts, src/storage/db.ts, src/core/reply-context.ts   (owned by S1 this plan)
+- src/agent/container-entry.ts, container/agent-package.json   (owned by S2 this plan)
+- src/config.ts   (owned by S4 this plan)
+- Not ours (avishai tsabari / Avishai-Tsabari, last 14 d): `src/agent/container-entry.ts`, `src/config.ts`, `src/storage/db.ts`, `container/agent-package.json` — see plan.md "Not ours"; lane cleared only by the Q1/Q2 answers. Re-check with `/f-ops-plan next p0-bugs-0922`.
+
+## Edges
+
+- start-after: none   (satisfied — `next` launched you)
+- merge-after: none
+
+## Overlap verdicts
+
+none
+
+> The command's preflight overlap check treats a hit whose branch AND every shared file match a verdict line as answered (c), prints one `PLAN-PASSED(overlap): …` line, and carries it into the final report. Any other hit prompts as usual. The merge-time check is never waived.
+
+## Decisions answered
+
+- none
+
+## Ledger rules
+
+- `docs/ROADMAP.md`: never edit. Your command writes `docs/pending-updates/subagent-parallel-mode-reports-api-errors-as-completed.md` as usual.
+- `docs/pending-verification.md`: append only under `## Plan p0-bugs-0922 / S3`.
+- Any goal's `decisions.md`: never edit; put decisions in your log's "Siblings must know".
+
+## Environment
+
+- Read `CLAUDE.md` → Worktree Setup and Gotchas before opening the worktree. Nothing from them is copied here. `bun install` once in the fresh worktree; the check gate is `bun run check` (every link runs to completion — read every line).
+
+## Report
+
+Your command's final report is appended to `docs/plans/p0-bugs-0922/sessions/S3.md` by its report step (path resolved through `.claude/scripts/resolve-docs-root.sh docs/plans/probe.md`). `work` verifies the append.

package/docs/plans/archive/2026-09-22-p0-bugs-0922/briefs/S4.md

@@ -0,0 +1,58 @@
+# Brief S4 — four boolean env keys turn off on a blank line
+
+**Plan**: docs/plans/p0-bugs-0922/plan.md   **Session**: S4   **Round**: 1   **Kind**: bug
+**Item slug**: four-boolean-env-keys-turn-off-on-a-blank-line
+
+> The only file this session reads from the plan. Self-contained: it never points at a sibling's brief. The `**Item slug**:` line is the key the execution commands grep for to detect that a brief is active — keep it exactly in that shape.
+
+## Launch
+
+/f-ops-plan work p0-bugs-0922 S4
+
+## Run
+
+/f-bug-fix four-boolean-env-keys-turn-off-on-a-blank-line
+
+## Scope notes
+
+The four keys: `telegramFormatEnabled`, `mediaEnabled`, `taskNotifyOnFailure` and `alertsEnabled`, all `booleanFromEnv.default(true)`. Reuse the strict parser `b6ecf07` added for `override_pi_system_prompt`: blank means unset, so YAML or the default applies; `true`/`false`/`1`/`0` (case-insensitive) is the value; anything else is a named startup error. Do **not** change the shared `booleanFromEnv` for its `default(false)` users; that is out of scope. Tests go in `tests/config.test.ts`: one table over 4 keys × {absent, blank, true, false, typo}. Negative-test by reverting one key to the lenient parser and confirming its row goes red. **Before landing**, check that no live `.env` on the box holds a typo'd value for these keys (key names only, never read values). The startup error would stop the service. **Deploy needs:** restart only.
+
+## Owned surfaces (predicted)
+
+- src/config.ts   (user-supplied, 2026-09-22)
+
+## Do not touch
+
+- src/core/storage-cleanup.ts, src/storage/db.ts, src/core/reply-context.ts   (owned by S1 this plan)
+- src/agent/container-entry.ts, container/agent-package.json   (owned by S2 this plan)
+- resources/pi-extensions/subagent/index.ts   (owned by S3 this plan)
+- Not ours (avishai tsabari / Avishai-Tsabari, last 14 d): `src/agent/container-entry.ts`, `src/config.ts`, `src/storage/db.ts`, `container/agent-package.json` — see plan.md "Not ours"; lane cleared only by the Q1/Q2 answers. Re-check with `/f-ops-plan next p0-bugs-0922`.
+
+## Edges
+
+- start-after: none   (satisfied — `next` launched you)
+- merge-after: none
+
+## Overlap verdicts
+
+none
+
+> The command's preflight overlap check treats a hit whose branch AND every shared file match a verdict line as answered (c), prints one `PLAN-PASSED(overlap): …` line, and carries it into the final report. Any other hit prompts as usual. The merge-time check is never waived.
+
+## Decisions answered
+
+- Q1: ours (owner, 2026-09-22) — `src/agent/container-entry.ts`, `src/config.ts`, `src/storage/db.ts` are this plan's lane.
+
+## Ledger rules
+
+- `docs/ROADMAP.md`: never edit. Your command writes `docs/pending-updates/four-boolean-env-keys-turn-off-on-a-blank-line.md` as usual.
+- `docs/pending-verification.md`: append only under `## Plan p0-bugs-0922 / S4`.
+- Any goal's `decisions.md`: never edit; put decisions in your log's "Siblings must know".
+
+## Environment
+
+- Read `CLAUDE.md` → Worktree Setup and Gotchas before opening the worktree. Nothing from them is copied here. `bun install` once in the fresh worktree; the check gate is `bun run check` (every link runs to completion — read every line).
+
+## Report
+
+Your command's final report is appended to `docs/plans/p0-bugs-0922/sessions/S4.md` by its report step (path resolved through `.claude/scripts/resolve-docs-root.sh docs/plans/probe.md`). `work` verifies the append.

package/docs/plans/archive/2026-09-22-p0-bugs-0922/decisions.md

@@ -0,0 +1,8 @@
+# Decisions — plan p0-bugs-0922
+
+> Answer in the `Answer` column. Everything blocked on a question stays blocked until this column is non-empty for it. `build` and `next` add rows; only you write answers. Delete nothing — an answered row is the audit trail.
+
+| ID | Blocks | Question | Options | Answer | Answered |
+|----|--------|----------|---------|--------|----------|
+| Q1 | S1, S2, S4 | Lane: does avishai tsabari own `src/` (3 files in 14d: `src/agent/container-entry.ts`, `src/config.ts`, `src/storage/db.ts`; last touch 2026-09-13)? | ours / theirs / split: … | ours — owner, relayed in session | 2026-09-22 |
+| Q2 | S2 | Lane: does Avishai-Tsabari own `container/agent-package.json` (1 file in 14d: PR #100, pi 0.85.1 bump, 2026-09-13)? S2's cheap fix edits only the prompt snippet. It touches this file only if it switches to a pi build whose `read` handles URLs. | ours / theirs / split: … | ours — owner, relayed in session | 2026-09-22 |

package/docs/plans/archive/2026-09-22-p0-bugs-0922/plan.md

@@ -0,0 +1,102 @@
+# Plan: p0-bugs-0922
+
+**Status**: Archived
+**Created**: 2026-09-22   **Last updated**: 2026-09-22
+**Landing mode**: pr
+**Max width**: 4
+**Inputs**: p0-bugs-0922 storage-cleanup-nulls-every-attachment-row-in-the-space system-prompt-advertises-url-capable-read-the-image-lacks subagent-parallel-mode-reports-api-errors-as-completed four-boolean-env-keys-turn-off-on-a-blank-line (from docs/notes/2026-09-22-open-items-priority-scan.md P0 #1, #3, #7)
+**Me**: saharw01, Saharw01@gmail.com (also 92163791+saharw01@users.noreply.github.com, avishailevi2@gmail.com)
+
+> Written by `/f-ops-plan` build (and `--rebuild`) and `archive` only. Nothing perishable lives here — no base sha, no other author's file list, no story counts; `next` measures those every time it runs. `plan-state.sh` parses the tables below by their header row, so keep the column order.
+
+> **Why this batch.** The bug half of the 2026-09-22 open-items scan's suggested round (P0 #1–3, #5–7). The ideas in that set (#2, #6, and #5's opt-out) are the sibling plan `p0-plan-0922`: this plan was split out on the owner's call (2026-09-22) so that four small fixes do not wait for their Round 0 planning sessions, the same split `bench-patterns-0915` / `bugs-0915` made. **Deploy:** S2 and S3 change baked files (`src/agent/container-entry.ts`, `resources/`), so the image needs one rebuild for both. S1 and S4 need only a restart.
+
+## Items
+
+| ID | Kind | Slug | Doc | Command | Round | Needs |
+|----|------|------|-----|---------|-------|-------|
+| S1 | bug | storage-cleanup-nulls-every-attachment-row-in-the-space | docs/bugs/storage-cleanup-nulls-every-attachment-row-in-the-space.md | /f-bug-fix storage-cleanup-nulls-every-attachment-row-in-the-space | 1 | Q1 |
+| S2 | bug | system-prompt-advertises-url-capable-read-the-image-lacks | docs/bugs/system-prompt-advertises-url-capable-read-the-image-lacks.md | /f-bug-fix system-prompt-advertises-url-capable-read-the-image-lacks | 1 | Q1, Q2 |
+| S3 | bug | subagent-parallel-mode-reports-api-errors-as-completed | docs/bugs/subagent-parallel-mode-reports-api-errors-as-completed.md | /f-bug-fix subagent-parallel-mode-reports-api-errors-as-completed | 1 | — |
+| S4 | bug | four-boolean-env-keys-turn-off-on-a-blank-line | docs/bugs/four-boolean-env-keys-turn-off-on-a-blank-line.md | /f-bug-fix four-boolean-env-keys-turn-off-on-a-blank-line | 1 | Q1 |
+
+> Kind ∈ `feature` | `bug` | `refactor` | `plan` (a Round 0 planning session) | `story` (from `--goal`; runs `/f-feature-dev`).
+> Needs = decision IDs from `decisions.md` that must be answered before this session may launch, or `—`.
+> IDs are never renumbered; `--drop` moves a row to Deferred, `--add` takes the next unused ID.
+
+## Edges
+
+| From | To | Kind | Basis |
+|------|----|------|-------|
+
+> None. No file is predicted by two sessions, the owner named no semantic dependency among these four (2026-09-22), and no other active plan's unlanded session predicts any of these files. The other active plans (`bench-patterns-0915`, `flat-0916`, `p0-plan-0922`) hold only planning sessions, and those have no surfaces.
+
+## Rounds
+
+### Round 0 — unblock
+
+| Session | Action | Clears |
+|---------|--------|--------|
+| — | owner answers decisions.md Q1–Q2 (lane questions) | S1, S2, S4 |
+
+### Round 1
+
+| Session | Item | Predicted surfaces (count) |
+|---------|------|----------------------------|
+| S1 | storage-cleanup-nulls-every-attachment-row-in-the-space | 3 |
+| S2 | system-prompt-advertises-url-capable-read-the-image-lacks | 2 |
+| S3 | subagent-parallel-mode-reports-api-errors-as-completed | 1 |
+| S4 | four-boolean-env-keys-turn-off-on-a-blank-line | 1 (user-supplied) |
+
+## Collision map
+
+> Whole files, predicted; `docs/` and infrastructure paths are excluded exactly as `overlap-check.sh` excludes them (1 such path ignored: `docs/ideas/gates-do-not-cover-resources-pi-extensions.md`, from S3).
+
+| File | Sessions |
+|------|----------|
+
+## Predicted surfaces
+
+### S1
+
+- src/core/storage-cleanup.ts
+- src/storage/db.ts
+- src/core/reply-context.ts
+
+### S2
+
+- src/agent/container-entry.ts
+- container/agent-package.json
+
+### S3
+
+- resources/pi-extensions/subagent/index.ts
+
+### S4
+
+- src/config.ts   (user-supplied, 2026-09-22)
+
+## Adjudicated overlaps
+
+> Pre-answers for the execution commands' preflight overlap prompt. Scoped to exact branch + file pairs on branches that belong to NO active plan (a hand-opened worktree, another contributor). A branch of a sibling plan's session is never listed here — it becomes a cross-plan edge instead.
+
+| Session | Branch (not in this plan) | Files | Verdict | Decided |
+|---------|---------------------------|-------|---------|---------|
+
+> The preflight for all four sessions came back clear on 2026-09-22 (no live worktrees besides the primary checkout).
+
+## Not ours
+
+> Last measurement (`git log --since=14.days`, authors not in `Me`). Re-measured on every `next`; this table is the snapshot the briefs were written from.
+
+| File / area | Author | Basis |
+|-------------|--------|-------|
+| src/agent/container-entry.ts | avishai tsabari | 0c5b980, 71f0073 (2026-09-11, bash-call bound, /dev/shm sizing) |
+| src/config.ts | avishai tsabari | eaf58ba (2026-09-13, /spaces conversations + link) |
+| src/storage/db.ts | avishai tsabari | eaf58ba (2026-09-13) |
+| container/agent-package.json | Avishai-Tsabari | cb94bba (2026-09-13, PR #100 pi 0.85.1 bump) |
+
+## Deferred
+
+| Item | Reason | Home |
+|------|--------|------|

package/docs/plans/archive/2026-09-22-p0-bugs-0922/sessions/S1.md

@@ -0,0 +1,29 @@
+## Launched — 2026-09-22T10:42Z
+Branch: worktree-fix-storage-cleanup-nulls-every-attachment-row-in-the-space
+
+## Report — 2026-09-22T10:55Z
+Outcome: blocked
+Delivered: —
+Blocked-on: owner approval to push dev (local-only rule, CLAUDE.local.md)
+Deploy needs: restart
+Files changed:
+- docs/debug/moderate/2026-09-22-storage-cleanup-nulls-every-attachment-row-in-the-space.md
+- docs/debug/storage-cleanup-nulls-every-attachment-row-in-the-space.md
+- docs/pending-verification.md
+- src/core/storage-cleanup.ts
+- src/storage/db.ts
+- tests/db.test.ts
+- tests/storage-cleanup.test.ts
+Siblings must know:
+- db.clearSpaceAttachments is gone; the replacement is db.clearAttachmentRefs(spaceId, inboxFileNames) -> { refsRemoved, rowsChanged, unparseable }. src/storage/db.ts is S1-edited; rebase before touching it.
+--- final report (verbatim) ---
+~~~
+S1 storage-cleanup-nulls-every-attachment-row-in-the-space — fixed, squash-merged to LOCAL dev as dd26ccf (not pushed; owner said "go" = local only).
+- src/core/storage-cleanup.ts: cleanDir returns deletedNames; sweep calls db.clearAttachmentRefs(spaceId, names) instead of the space-wide clear.
+- src/storage/db.ts: clearSpaceAttachments -> clearAttachmentRefs (per ref, basename + parent `inbox`, one transaction, unparseable rows kept + counted).
+- tests: real-Db regression (old + fresh file, mixed row) + db unit tests; negative-tested (space-wide predicate -> 2 red).
+- Gate: 11/12 PASS; `test` red only on feed-watch-scorecard, identical on dev, filed as docs/bugs/feed-watch-scorecard-cli-test-pins-now-but-the-script-reads-the-real-clock.md.
+- Post-mortem: docs/debug/moderate/2026-09-22-storage-cleanup-nulls-every-attachment-row-in-the-space.md
+- MANUAL-VERIFY under "## Plan p0-bugs-0922 / S1" in docs/pending-verification.md.
+- Deploy: restart only. Residuals: no backfill of already-nulled rows; /media/purge leaves refs (consistent, noted).
+~~~

package/docs/plans/archive/2026-09-22-p0-bugs-0922/sessions/S2.md

@@ -0,0 +1,29 @@
+## Launched — 2026-09-22T10:43Z
+Branch: worktree-fix-system-prompt-advertises-url-capable-read-the-image-lacks
+
+## Report — 2026-09-22T11:02Z
+Outcome: delivered
+Delivered: —
+Blocked-on: —
+Deploy needs: rebuild
+Files changed:
+- docs/debug/moderate/2026-09-22-system-prompt-advertises-url-capable-read-the-image-lacks.md
+- docs/pending-verification.md
+- src/agent/container-entry.ts
+- tests/core-tool-snippets.test.ts
+- tests/fixtures/release-gate/regression-map.json
+Siblings must know:
+- A post-mortem archived under docs/debug/** now needs a `Gate check:` line or an entry in tests/fixtures/release-gate/regression-map.json, or tests/release-gate-map.test.ts goes red. S1's post-mortem (moderate/2026-09-22-storage-cleanup-nulls-every-attachment-row-in-the-space.md) has neither and is red on local dev.
+- tests/feed-watch-scorecard.test.ts is red on dev regardless of change (fixed NOW vs the script's real clock); filed as docs/bugs/feed-watch-scorecard-cli-test-pins-now-but-the-script-reads-the-real-clock.md (uncommitted).
+- src/agent/container-entry.ts exports CORE_TOOL_SNIPPETS.
+--- final report (verbatim) ---
+~~~
+Fixed: the override-mode system prompt listed `read` as "a file or URL" with web-page conversion, copied from an upstream pi; the pinned @earendil-works/pi-coding-agent 0.85.1 read opens local paths only.
+- src/agent/container-entry.ts: tool list is now exported CORE_TOOL_SNIPPETS; read = installed description's opening + "Local paths only — it cannot open a URL; fetch web pages with `curl` via bash."
+- tests/core-tool-snippets.test.ts: each summary must be a prefix of the installed tool's description; image pin == root pin and installed version satisfies it. Negative-tested (old text -> 2 red).
+- Post-mortem: docs/debug/moderate/2026-09-22-system-prompt-advertises-url-capable-read-the-image-lacks.md (+ regression-map entry, category code).
+- Landed as e73faa3 on LOCAL dev, squash; NOT pushed (CLAUDE.local.md push gate; owner chose local squash). Promotion commit de5fb69 also unpushed.
+- Gate after rebase: 11/12 links pass; test 5735 pass / 9 skip / 3 fail before the map entry -> remaining 2 fails are not S2's (S1's post-mortem missing from regression-map; pre-existing feed-watch-scorecard date bug).
+- Deploy: rebuild (container-entry.ts is baked) — container/build.sh then restart in a gap; verify by grepping the image (pending-verification, Plan p0-bugs-0922 / S2). One rebuild covers S3 too.
+- ARCHITECTURE.md: no change needed.
+~~~

package/docs/plans/archive/2026-09-22-p0-bugs-0922/sessions/S3.md

@@ -0,0 +1,30 @@
+## Launched — 2026-09-22T10:43Z
+Branch: worktree-fix-subagent-parallel-mode-reports-api-errors-as-completed
+
+## Report — 2026-09-22T13:05Z
+Outcome: delivered
+Delivered: —
+Blocked-on: —
+Deploy needs: rebuild
+Files changed:
+- docs/debug/minor/2026-09-22-subagent-parallel-mode-reports-api-errors-as-completed.md
+- docs/debug/subagent-parallel-mode-reports-api-errors-as-completed.md
+- docs/pending-verification.md
+- resources/pi-extensions/subagent/index.ts
+- resources/pi-extensions/subagent/outcome.ts
+- tests/subagent-outcome.test.ts
+Siblings must know:
+- none (resources/pi-extensions/subagent/ only; new module outcome.ts exports isSubagentError + RUNNING_EXIT_CODE)
+--- final report (verbatim) ---
+~~~
+S3 — subagent-parallel-mode-reports-api-errors-as-completed: fixed, squashed onto LOCAL dev as 78bfe74 (not pushed — owner chose "squash onto local dev"; push needs approval per CLAUDE.local.md).
+
+Fix: resources/pi-extensions/subagent/outcome.ts adds isSubagentError (exitCode !== 0 || stopReason error/aborted); all eleven verdict sites in index.ts use it — parallel successCount + completed/failed label, chain/single behaviour, and every ✓/✗ counter and icon (the chain display was also exit-code-only, beyond the report's four sites). A failed parallel leg now previews its error text instead of "(no output)".
+Test: tests/subagent-outcome.test.ts — predicate units + end-to-end execute/renderResult for single, chain and parallel against a fake pi on PATH that exits 0 with stopReason "error". Negative-tested: each of the 11 sites reverted alone turns the suite red.
+Checklist item "resources/ inside typecheck + lint": already done (typecheck:resources + lint both cover resources/) — verified and ticked, not redone.
+Gate (bun run check, WSL worktree): 11/12 links PASS; test 5737 pass / 9 skip / 1 fail — the failure is feed-watch-scorecard "prints the table…", identical on untouched dev, already filed (docs/bugs/feed-watch-scorecard-cli-test-pins-now-but-the-script-reads-the-real-clock.md).
+Rebase: pending-verification.md conflicted with S1/S2's appended sections; resolved keeping all three.
+Post-mortem: docs/debug/minor/2026-09-22-subagent-parallel-mode-reports-api-errors-as-completed.md (residuals: streaming parallel view counts a running task as done; parallel never sets isError even when all legs fail).
+Deploy needs: image rebuild — resources/ is baked. Verification lines under "## Plan p0-bugs-0922 / S3" in docs/pending-verification.md.
+ARCHITECTURE.md: no update needed.
+~~~

package/docs/plans/archive/2026-09-22-p0-bugs-0922/sessions/S4.md

@@ -0,0 +1,32 @@
+## Launched — 2026-09-22T10:44Z
+Branch: worktree-fix-four-boolean-env-keys-turn-off-on-a-blank-line
+
+## Report — 2026-09-22T13:10Z
+Outcome: blocked
+Delivered: —
+Blocked-on: owner approval to push dev (local-only rule, CLAUDE.local.md)
+Deploy needs: restart
+Files changed:
+- docs/debug/four-boolean-env-keys-turn-off-on-a-blank-line.md
+- docs/debug/minor/2026-09-22-four-boolean-env-keys-turn-off-on-a-blank-line.md
+- docs/pending-verification.md
+- src/config.ts
+- tests/config.test.ts
+- tests/fixtures/release-gate/regression-map.json
+Siblings must know:
+- src/config.ts now exports nothing new; strictBooleanFromEnv(configKey, yamlKey, default) is module-private and derives the env name from configKeyEnvNames().
+- tests/release-gate-map.test.ts is red on dev for S1's and S3's post-mortems (moderate/2026-09-22-storage-cleanup-nulls-every-attachment-row-in-the-space.md, minor/2026-09-22-subagent-parallel-mode-reports-api-errors-as-completed.md): each needs a Gate check: line or an entry in tests/fixtures/release-gate/regression-map.json. S4 added only its own entry.
+--- final report (verbatim) ---
+~~~
+S4 — four-boolean-env-keys-turn-off-on-a-blank-line: fixed, squash-merged to local dev as 722d697, NOT pushed.
+
+- src/config.ts: override_pi_system_prompt's strict parser generalised to strictBooleanFromEnv; telegramFormatEnabled, mediaEnabled, taskNotifyOnFailure, alertsEnabled now use it (blank = default true; true/false/1/0 case-insensitive; anything else = startup error naming YAML key + MERCURY_* var, the name derived from config-file's map).
+- tests/config.test.ts: 48-row table (4 keys x absent/blank/3 true/3 false/4 typos); negative test: alertsEnabled on the lenient parser -> 5/12 of its rows red.
+- tests/fixtures/release-gate/regression-map.json: allowlist entry for this post-mortem (category code).
+- docs/pending-verification.md: one MANUAL-VERIFY line under "## Plan p0-bugs-0922 / S4".
+- Post-mortem: docs/debug/minor/2026-09-22-four-boolean-env-keys-turn-off-on-a-blank-line.md
+- Gate on the rebased tree: 11/12 links PASS; test red on (a) feed-watch-scorecard clock-dependent test, pre-existing and filed, fails identically with this fix reverted; (b) release-gate-map rows for S1 and S3 post-mortems (not this session's).
+- Box: none of the four keys is set in ~/whatsapp-bot/.env or the user unit (key-name check only), so the new startup error cannot fire on restart.
+- Residual: a blank env line still shadows a mercury.yaml value with the default (config-file merge; same as b6ecf07), recorded in the post-mortem.
+- Architecture: no ARCHITECTURE.md change needed.
+~~~

package/docs/plans/bench-patterns-0915/briefs/S10.md

@@ -0,0 +1,72 @@
+# Brief S10 — feed-watch serves fixture files in shadow mode (M1.7)
+
+**Plan**: docs/plans/bench-patterns-0915/plan.md   **Session**: S10   **Round**: 1   **Kind**: feature
+**Item slug**: shadow-feed-fixtures
+
+> The only file this session reads from the plan. Self-contained: it never points at a sibling's brief. The `**Item slug**:` line is the key the execution commands grep for to detect that a brief is active — keep it exactly in that shape.
+
+## Launch
+
+/f-ops-plan work bench-patterns-0915 S10
+
+## Run
+
+/f-feature-dev shadow-feed-fixtures
+
+## Scope notes
+
+Follow the doc's six-step sequence. The fixture replaces only the transport; everything after the response stays the real code (Non-Negotiable Rules). For Step 4, read the configured source keys **read-only from the latest shadow snapshot's DB**, never from the live DB and never from `.env`, and compute them with `expandSources`. Do not type them by hand.
+
+The doc's Open Question about the stale gnews bug doc is not yours to close. Mention it in "Siblings must know".
+
+Until M1.3c (`shadow-own-credential`) is deployed, the shadow shares the live refresh-token chain. Any live fixture rehearsal you run is short and watched.
+
+**Deploy needs: restart.** feed-watch reaches the box as the installed extension snapshot, so verify against `~/whatsapp-bot/.mercury/extensions/feed-watch/`, never the repo. Gap-check before the restart.
+
+## Owned surfaces (predicted)
+
+- examples/extensions/feed-watch/fixtures.ts
+- examples/extensions/feed-watch/index.ts
+- tests/feed-watch-fixtures.test.ts
+- tests/fixtures/feed-watch/shadow/README.md
+- tests/fixtures/feed-watch/shadow/feeds
+- tests/fixtures/feed-watch/shadow/feeds/*
+- docs/pending-verification.md
+
+## Do not touch
+
+- src/cli/mercury.ts, src/config.ts, src/ops/shadow-root.ts, src/ops/shadow-service.ts, src/ops/shadow-snapshot.ts, tests/shadow-login.test.ts, tests/shadow-mode.test.ts, tests/shadow-root.test.ts, tests/shadow-service-cli.test.ts, tests/shadow-service.test.ts, tests/shadow-snapshot.test.ts, docs/configuration.md, docs/live-testing.md   (owned by S9 this plan)
+- package.json, scripts/ (every file), tests/check-ratchets.test.ts, tests/check-sync-calls.test.ts   (owned by S11 this plan)
+- tests/helpers/, tests/inventories/, tests/inventory-*.test.ts, tests/source-scan.test.ts, tests/container-copy-list-closure.test.ts, tests/env-claims-declare-hostonly.test.ts, tests/no-module-eviction.test.ts, tests/release-gate-map.test.ts   (owned by S12 this plan)
+- examples/extensions/feed-watch/feeds.ts, watch.ts, config.ts: read-only for this item (the doc says so).
+- src/core/routes/shadow.ts, tests/shadow-route.test.ts, src/core/simulate-ingress.ts, src/smoke/cli.ts, src/bridges/, src/core/conversation.ts: named in the doc's Non-Negotiable Rules.
+- Not ours: none intersect this session's surfaces (measured 2026-09-23). Re-check with `/f-ops-plan next bench-patterns-0915`.
+
+## Edges
+
+- start-after: none   (satisfied — `next` launched you)
+- merge-after: none
+
+## Overlap verdicts
+
+none
+
+> The command's preflight overlap check treats a hit whose branch AND every shared file match a verdict line as answered (c), prints one `PLAN-PASSED(overlap): …` line, and carries it into the final report. Any other hit prompts as usual. The merge-time check is never waived.
+
+## Decisions answered
+
+- Q1: yes. The bench order is M1.3c → M1.7 → M2.4 (owner, 2026-09-23). There is no landing order between you and S9 (owner: no semantic edges).
+
+## Ledger rules
+
+- `docs/ROADMAP.md`: never edit. Your command writes `docs/pending-updates/shadow-feed-fixtures.md` as usual.
+- `docs/pending-verification.md`: append only under `## Plan bench-patterns-0915 / S10`. S9 appends under its own heading in the same file, so rebase, don't overwrite.
+- Any goal's `decisions.md`: never edit; put decisions in your log's "Siblings must know".
+
+## Environment
+
+- Read `CLAUDE.md` → Worktree Setup and Gotchas before opening the worktree. Nothing from them is copied here. Run `bun install` once in the fresh worktree. The check gate is `bun run check`: every link runs to completion, so read every line.
+
+## Report
+
+Your command's final report is appended to `docs/plans/bench-patterns-0915/sessions/S10.md` by its report step (path resolved through `.claude/scripts/resolve-docs-root.sh docs/plans/probe.md`). `work` verifies the append.

package/docs/plans/bench-patterns-0915/briefs/S11.md

@@ -0,0 +1,95 @@
+# Brief S11 — ratchet the four "silence" shapes the summaries named
+
+**Plan**: docs/plans/bench-patterns-0915/plan.md   **Session**: S11   **Round**: 1   **Kind**: feature
+**Item slug**: ratchet-shapes-the-summaries-named
+
+> The only file this session reads from the plan. Self-contained: it never points at a sibling's brief. The `**Item slug**:` line is the key the execution commands grep for to detect that a brief is active — keep it exactly in that shape.
+
+## Launch
+
+/f-ops-plan work bench-patterns-0915 S11
+
+## Run
+
+/f-feature-dev ratchet-shapes-the-summaries-named
+
+## Scope notes
+
+**First step, before the worktree and before any detector** (owner decision Q9/OQ4 and Q10): file both candidate bugs with `/f-bug-report`, then put their two slugs in your log's "Siblings must know":
+
+- (a) `src/core/handler.ts:315`: the `denied` reply is an `await bridge.sendReply` outside any `try`.
+- (b) `src/core/media.ts:233`: with no `content-length`, the whole body is buffered by `arrayBuffer()` before the size check.
+
+File them, never fix them. Their fixes later lower your baselines.
+
+The owner accepted all five Open-Question recommendations (Q9):
+
+- OQ1: ratchet D2 at its measured count; no sweep first.
+- OQ2: skip `src/bridges/` and `src/adapters/`, and say why in the script header.
+- OQ3: a zero-dependency lexer.
+- OQ4: file the two bugs, as above.
+- OQ5: no suppression marker.
+
+`check-silent-catch.ts` and `check-sync-calls.ts` stay byte-identical. They are predicted only because the doc names them as patterns. Re-measure the baselines on your branch's own tree (Step 3); the planning table is not the baseline.
+
+**Deploy needs: none** (scripts and tests only).
+
+## Owned surfaces (predicted)
+
+- package.json
+- scripts/absent-allows-baseline.json
+- scripts/bare-delivery-await-baseline.json
+- scripts/check-absent-allows.ts
+- scripts/check-bare-delivery-await.ts
+- scripts/check-early-return-no-log.ts
+- scripts/check-silent-catch.ts
+- scripts/check-sync-calls.ts
+- scripts/check-unbounded-race.ts
+- scripts/check.ts
+- scripts/early-return-no-log-baseline.json
+- scripts/lib/ratchet.ts
+- scripts/silent-catch-baseline.json
+- scripts/unbounded-race-baseline.json
+- tests/check-ratchets.test.ts
+- tests/check-sync-calls.test.ts
+- CLAUDE.md
+
+## Do not touch
+
+- src/cli/mercury.ts, src/config.ts, src/ops/shadow-root.ts, src/ops/shadow-service.ts, src/ops/shadow-snapshot.ts, tests/shadow-*.test.ts, docs/configuration.md, docs/live-testing.md   (owned by S9 this plan)
+- examples/extensions/feed-watch/fixtures.ts, examples/extensions/feed-watch/index.ts, tests/feed-watch-fixtures.test.ts, tests/fixtures/feed-watch/shadow/   (owned by S10 this plan)
+- tests/helpers/, tests/inventories/, tests/inventory-*.test.ts, tests/source-scan.test.ts, tests/container-copy-list-closure.test.ts, tests/env-claims-declare-hostonly.test.ts, tests/no-module-eviction.test.ts, tests/release-gate-map.test.ts   (owned by S12 this plan)
+- Everything under `src/` and `examples/`: scanned, never edited. A hit there goes into a baseline or a bug report. This includes `src/core/handler.ts`, `src/core/media.ts` and `src/core/simulate-ingress.ts`, and the files plans `p0-plan-0922` (S6–S8) and `sensitive-bugs-0922` are changing.
+- `CLAUDE.md`: S12 also edits it (the `pi`-spawn gotcha). Add your one Gotchas line and nothing else, and rebase over theirs.
+- Not ours (avishai tsabari / dependabot, last 14 d): `package.json`, from version bumps and dependency PRs. The lane is cleared by Q12 (ours). Re-check with `/f-ops-plan next bench-patterns-0915`.
+
+## Edges
+
+- start-after: none   (satisfied — `next` launched you)
+- merge-after: none
+
+## Overlap verdicts
+
+none
+
+> The command's preflight overlap check treats a hit whose branch AND every shared file match a verdict line as answered (c), prints one `PLAN-PASSED(overlap): …` line, and carries it into the final report. Any other hit prompts as usual. The merge-time check is never waived.
+
+## Decisions answered
+
+- Q9: accept all five Open-Question recommendations (owner, 2026-09-23).
+- Q10: the two bug reports are your first step, not a launch gate (owner, 2026-09-24).
+- Q12: ours (owner, 2026-09-23).
+
+## Ledger rules
+
+- `docs/ROADMAP.md`: never edit. Your command writes `docs/pending-updates/ratchet-shapes-the-summaries-named.md` as usual.
+- `docs/pending-verification.md`: append only under `## Plan bench-patterns-0915 / S11`.
+- Any goal's `decisions.md`: never edit; put decisions in your log's "Siblings must know".
+
+## Environment
+
+- Read `CLAUDE.md` → Worktree Setup and Gotchas before opening the worktree. Nothing from them is copied here. Run `bun install` once in the fresh worktree. The check gate is `bun run check`: every link runs to completion, so read every line. Lint is proven only by `~/check-mercury.sh check <worktree>` in WSL (CLAUDE.local.md).
+
+## Report
+
+Your command's final report is appended to `docs/plans/bench-patterns-0915/sessions/S11.md` by its report step (path resolved through `.claude/scripts/resolve-docs-root.sh docs/plans/probe.md`). `work` verifies the append.