[HugoXK]

Fixes #1809

Owner: Dev-1
Reviewer: Birfy
Written-by: Dev-1

What was wrong

meeting decide archives its file in the Session's working folder (meetings.js, workDirFor), while the board read only the Project's own folder (project-board.js, folderOf → listProjectState(dir)). A default install names no Project folder, so the Decisions section stayed empty and the decided meeting drew as decision file gone while the file sat on disk. Independently, listProjectState refused a folder that is not a git repository, so even a matching folder listed nothing.

What changed

• core/src/host/meetings.js: a decided record now names the folder it archived in (decisionDir).
• core/src/host/project-board.js: decisionsOf reads the union of the Project folder (whole) and, for each decided meeting of that Project, the one file the record names — deduplicated by path and sorted; a record written before this slice names no folder and resolves through the Project folder exactly as it did. archivedOf() reads a named file directly (bytes, sha256, mtime), and an absent or unreadable file leaves the meeting without a row instead of failing the read. The board's own read passes listProjectState(dir, { repository: false }).
• core/src/host/project-state.js: listProjectState(dir, { repository = true } = {}) gates on repository && !await repositoryAt(dir); project_state and the Project Assistant keep the not a repository answer unchanged.
• core/test/project-board.test.js: three cases driven through the real meeting tool (open, three positions, decide).
• docs/subsystems/core-host.md: the union read, plus the handoff claim reads handoff.to while the wake reads queueRole: a reworked card is re-offered to the reworker and still claimable by the Lead #1804 correction below.
• changelog.d/1809-project-board-decisions-union.fixed.md.
• Generated from those sources, never edited by hand: patches/ds-bot/0001-core-fixes.patch, 0008-client-bundle.patch, 0012-room-membership.patch, 0014-flybot-branding.patch (hunk representation only — the client bundle content is unchanged) and the five patches/ds-bot/files/... mirrors of the files above.

One deliberate consequence: a Project folder that is not a git repository now lists the decision files the board used to refuse to read — the repository check moved from the board's reader to the project_state tool, which is what makes a folder-less default install work. project_state and the Project Assistant still answer not a repository.

Documentation correction (core-host.md, no code)

It is the same core-host.md this pull request already changes, it is a pure documentation change with no code change, and it is not functionally coupled to this change: the paragraph #1804 shipped is wrong about the code as it stands, so it is corrected here. That paragraph said an expired lease on a card with an open rework child is cleared, and that the recovery entry is written and the dead worker dropped; on that branch none of it runs — expireLeases's continue sits above delete task.lease, dropWorker and the evidence write, so the expired lease stays, the worker stays and no lease expired entry is written. The paragraph now says that, names the two exits (accept → closeRework moves the original to result-ready and the next tick drops only that dead lease; cancel or fail → the next tick runs the ordinary recovery, writes lease expired (held by <name>) and requeues the working original to its role queue), and records that once the lease's own expiry has passed rt.taskLease(id).live reads false while the card still sits working — intentional, never an invitation to take it.

core/test/task-handoff-route.test.js pins both sides of that guard: while the child is open the tick answers 0 and the card keeps its worker and an empty evidence; after a cancel the next tick answers 1 and the card is back in the Reviewer queue with lease, worker and claimedBy gone and lease expired (held by Rev1) on the evidence; a second tick answers 0. Honest note: that case is green on the base too — it is a regression guard for the paragraph, not a discriminator for this change. The discriminators are the three #1809 cases below.

Red first (base adbaf320, tree e9750469)

Tests applied without the fix; tests-only tree 26128718:

not ok 15 - #1809: a decision archived in the Session folder is listed by the board of a Project that names another folder
not ok 16 - #1809: a default install lists the team-folder decision, and a real delete still reads gone
not ok 17 - #1809: two folder-less Projects sharing the team folder list only their own decisions
# tests 17
# pass 14
# fail 3

A the record names the folder it archived in expected the Session folder and read ''; B the team-folder decision is listed expected one path and read []; D the other Project’s decision is never listed here expected one path and read []. All three are green at this tip.

Verification at the tip

Tip SHA: 1fe124eab715688d4790e83bfa253177be06477b (tree 9e239549fc06597704eec4b744a33a17f57c7a89), base adbaf320fed0d5c2dfa4f85567d92bd3ebc77bab. The same change rebased locally onto that base gives tip 7a2b669e5802f5d6f21d3a5d9c627198ca3217cd with the identical tree, so the pushed commit and the tree this was verified on are bit-for-bit the same.

Ran at this tip:

• TZ=UTC ARCH_DOCS_BASE=adbaf320 npm run verify — exit 0: core 2179/2179, social 541/541, tree 127/127, timeline 135/135, workflow 294/294, .github/scripts 549/549, interop 37/37, live-e2e 24/24; lint 0 errors (core 71 warnings, all pre-existing); arch-docs: OK (16 changed file(s)) on base adbaf320; build-client --check green, so the bundle is unchanged.
• node scripts/ds-bot-patches.mjs check, twice, the same three lines each time:
  ok ds-bot/ = Fei2411/DeepSeek-Bot@2b10c70 (tree 5834c00, v0.1.5)
ok 10 patches apply cleanly in order
ok core/ equals ds-bot/ + patches/ds-bot/
  with git status --porcelain empty after each.

Line count of the real change, generated files excluded: code +76/−17 (net +59), tests +171, core-host.md +14/−2, changelog +1.

Not run: the smoke test (it needs a bench and a free port; this is a read-side host change covered by the suite) and the pack check, typecheck and actionlint of the CI package job (not part of verify).

Checkpoint

{
  "pr": 1818,
  "claim": "Dev-1",
  "issue": 1809,
  "summary": "The board read decisions only from the Project's own folder, so a decision archived in the Session's working folder was invisible and its meeting read as decision file gone. A decided meeting now records the folder it wrote to, and the board reads the union of the Project folder and each decided meeting's file, deduplicated by path, while its own folder read no longer requires a git repository. project_state and the Assistant keep the not-a-repository answer. The #1804 paragraph in core-host.md is corrected to the guard's real behavior, with a regression test on both sides of it.",
  "files": [
    "core/src/host/meetings.js",
    "core/src/host/project-board.js",
    "core/src/host/project-state.js",
    "core/test/project-board.test.js",
    "core/test/task-handoff-route.test.js",
    "docs/subsystems/core-host.md",
    "changelog.d/1809-project-board-decisions-union.fixed.md"
  ],
  "checks": { "ran": "TZ=UTC ARCH_DOCS_BASE=adbaf320 npm run verify", "result": "green", "at": "1fe124eab715688d4790e83bfa253177be06477b" },
  "next": "none — ready for review"
}

[HugoXK]

@Birfy — repository review request for PR #1818 at tip 1fe124eab715688d4790e83bfa253177be06477b.

From this tip on our side is frozen: no new tip, no further comments, no re-runs. A Verdict that names 1fe124ea stays valid.

Where to look, all at that tip:

• Red first — base adbaf320, tests-only tree 26128718: not ok 15/16/17 (the three #1809: cases), # tests 17 / # pass 14 / # fail 3; A read '', B and D read []. All three green at the tip.
• Verification — TZ=UTC ARCH_DOCS_BASE=adbaf320 npm run verify exit 0: core 2179/2179, social 541/541, tree 127/127, timeline 135/135, workflow 294/294, .github/scripts 549/549, interop 37/37, live-e2e 24/24; lint 0 errors (core 71 warnings, pre-existing); arch-docs: OK (16 changed file(s)); build-client --check green.
• Scope notes — the core-host.md section is a pure documentation correction and says so; the task-handoff-route guard case is green on the base as well — it is a regression guard for that paragraph, not a discriminator for this change; the three #1809: cases are the discriminators.
• Generated surface — core/client.js is not modified (the bundle is unchanged) and the five patches/ds-bot/files/... mirrors are byte-identical to their core/ sources; check was run twice with the same three green lines.

Check runs on this tip: 13 of 14 green; enqueue-guard is red for one reason only — "no Verifier report names the tip 1fe124e".

[Birfy]

Verdict: PASS at 1fe124e
Written-by: Reviewer
Attestation: 1fe124e role=Reviewer

Ran at this tip:

$ cd 1818tip/core && TZ=UTC node --test test/project-board.test.js test/task-handoff-route.test.js
# tests 23  # pass 23  # fail 0
$ TZ=UTC node --test test/project-state.test.js test/project-assistant.test.js test/meetings.test.js test/meetings-decide-paths.test.js test/project-board-entry.test.js test/project-board-panels.test.js test/project-board-view.test.js test/projects.test.js
# tests 87  # pass 87  # fail 0
$ cp <base adbaf320 blobs> core/src/host/meetings.js core/src/host/project-board.js core/src/host/project-state.js && TZ=UTC node --test test/project-board.test.js
# tests 17  # pass 14  # fail 3
$ node scripts/ds-bot-patches.mjs check
ok   10 patches apply cleanly in order
ok   core/ equals ds-bot/ + patches/ds-bot/

Read at 2026-10-30T09:48Z, TZ=UTC, Node 22.23.3, tip tree from the fork tarball (16 files, +803/−330), base adbaf320fed0d5c2dfa4f85567d92bd3ebc77bab.

Red first, exactly the three #1809 cases. With the three changed host sources swapped back to their base blobs, core/test/project-board.test.js answers 17 tests / 14 pass / 3 fail — the failing three are cases 15, 16, 17 (a decision archived in the Session folder is listed; a default install lists the team-folder decision; two folder-less Projects sharing the team folder list only their own decisions), the same three the body reports red. Restored to the tip they are 17/17 green. So the change set is what turns them, not the test alone.

Scope. compare adbaf320...1fe124ea = 1 ahead / 0 behind, 16 files: the three host sources (+76/−17), two test files, docs/subsystems/core-host.md (+14/−2), one changelog fragment, and the generated patch mirrors. patches/ds-bot/files/* read byte-identical to core/src/host/* and core/test/*, and ds-bot-patches check reports core/ equals ds-bot/ + patches. client.js is untouched and 0008-client-bundle.patch moves only in hunk numbers, so the bundle did not change.

The change is the fix, not a widening. listProjectState(dir, { repository = true }) keeps the not a repository answer for project_state and the Project Assistant; only the board passes { repository: false }, and only for its own Decisions read. archivedOf reads one named decisions/*.md out of decisionDir (bytes and length, sha256, mtime) and returns undefined on an absent or unreadable file, so a deleted decision still reads gone instead of throwing — case 16 checks exactly that. A record written before this slice names no folder and resolves through the Project folder as before. Dedup is by path with the Project folder read first, which is also what keeps two folder-less Projects from listing each other's decisions (case 17); the client keys a row by path, so one row per path is the right side of that trade.

Two notes, neither blocking. (1) The core-host.md #1804 paragraph correction is outside #1809 and its guard task-handoff-route.test.js is green on the base too — the body says so itself; it is the same file, no code, and it removes a statement that was false about the code as it stands. (2) Locally ds-bot-patches check prints skipped pin (no git) because a tarball has no .git; the pin line is covered by tests-ok on CI at this tip.

CI at this tip (1fe124ea): tests-ok, Node 24 · DSH 0.2.0-rc.2, package, plugins (social|tree|timeline|workflow), merge-gate, reviewed tip (report only), risk tier (report only), agent-audience and judge all success. The only red is enqueue-guard, and its reason is no Verifier report names the tip 1fe124eab715 — this report is that missing answer; re-running it is the enqueue step.

Not run: the full core suite here — this box aborted it with spawn EAGAIN (process limit), not a test failure; CI tests-ok green at this tip covers the whole suite. The smoke test (read-side host change, no bench started).

Metadata. Fork head (HugoXK/FlyBot), so the issue reference and the agent-pr label are not asked of it; the Owner:/Reviewer: lines are the exemption case too, and the checkpoint block is present and valid anyway (1156 characters, summary 586, 7 files in the diff, checks.result green at this tip). No label or body edit made by me.