[HugoXK]

Fixes #1758

Three defects in the digest text a Project owner receives, all in
core/src/host/project-assistant.js: a quiet window compared as wall-clock text against a due
instant, silent truncation past five waiting decisions, and cards that wait on nobody listed as
waiting on the owner.

Base: 7f4aa2a70766a7292cf5474ec8a7bbc364f5d1ea (the tip after #1747 merged; that merge touches no
file this PR touches).

Head: f01a17c6c9a52c0e3e047447292960f22a12f9de (tree 8031b5b95085ead0d199dea0b5d52e93e9b72e18;
the locally verified commit 8a67f9a0e80e6c556810dd944c5ec09d3384dc4f has the same tree — the push
helper rebuilt the commit).

What changed

• P-194 — quiet window vs. due instant. The due time is 09:00 Asia/Shanghai, an instant; a
  quiet window is wall clock in the owner's own zone. Comparing the string 09:00 with
  inQuietHours compared two different zones' wall clocks, so an owner outside +08:00 was dropped
  on the wrong day. The due instant is now computed (Date.parse(day + 'T09:00:00+08:00'), read
  back with clockIn(quiet.zone, …)) and compared as a clock in the owner's zone. The held label
  uses the instant the digest was actually held at, and the drop reason names the due clock rather
  than the literal 09:00.
• P-197 — overflow line. The personal block stays capped at five; when more than five cards
  wait, a sixth line says … and N more waiting on you. Five or fewer print exactly the bytes they
  printed before.
• P-195 — only what waits on the reader. The personal block now holds blocked ("needs your
  call") and result-ready ("ready for your acceptance") cards, read from the card's own status. A
  card still open, claimed or working waits on nobody and appears only in the shared
  in-progress section. A row that carries no status keeps the older answer — absence is not
  evidence that the card is moving.

Red first

Red first drives the digest through rt.projectAssistant.setOwner(...) directly: that endpoint has
no channel case (core/src/host/store.js:608 handles project-assistant-owner, but
core/src/host/channel.js:670-677 lists only the eight project-* endpoints, so the request falls
to default and fails Unknown endpoint / bot/not-found at :716). No tool, hook, CLI or
schedule path reaches it either, so no product-level path exists today; the window matters once
P-198 wires it.

What is visible today, in two classes:

• P-194 is a P-198 prerequisite: without a quiet record, quietOf (project-assistant.js:294)
  yields {on:false}, so the coversDue / label / reason paths cannot fire.
• P-197 and P-195 write the digest text today but have no live producer: the running instance
  has projects: 0, and project-create has no client caller and no UI path, so no Project exists
  to produce a digest.

On base 7f4aa2a7 with only the source reverted (# tests 22 / # pass 16 / # fail 6), the six
failures are:

not ok 10 - a quiet window holds the digest and delivers it once, labelled, when it ends
not ok 12 - #P-194: a quiet window in the owner's zone is compared with the due instant, not with 09:00
not ok 13 - #P-194: a window that really covers the due instant still drops the digest
not ok 14 - #P-197: more than five decisions say how many are still waiting
not ok 16 - #P-195: only the cards that wait on the reader reach the personal block
not ok 20 - a card lists for the owner of the Bot that owns it, through the real task and board path

With the fix: 22/22 at TZ=UTC and 22/22 at TZ=Asia/Shanghai.

The process time zone changes nothing; the owner's zone does. Same probe, base source vs. this
branch, owner quiet 09:00–11:00 @ Asia/Shanghai — the reason is byte-identical in both process
zones: quiet hours 09:00–11:00 cover 09:00, so nothing is delivered — the expected outcome. Owner
quiet 08:00–10:00 @ UTC, 01:00Z: base drops the digest with cover 09:00; this branch delivers
once and carries no reason.

Two existing tests changed

• Test 10 now expects held from 08:00 instead of 09:00. The old string was
  held from 09:00 to 08:45 (base :325) — it starts after it ends, because 09:00 is exactly
  the DIGEST_TIME constant (project-assistant.js:24) the old code printed in the label
  (:375) and compared in inQuietHours(quiet, DIGEST_TIME) (:365): the old expectation
  asserted the defect itself. The label now starts at the instant the digest was actually held
  (clockIn(quiet.zone, held.at ?? at)), which is 08:00 in UTC.
• Test 20's old fixture used only task-create (default status: 'open') and still expected
  two cards in the personal block, which P-195 no longer allows. The new fixture drives both cards
  through task-accept → task-start → task-wait-input to blocked, and adds a working card with
  the a card the owner's Bot is still working on waits on nobody negative assertion. The four
  original title assertions are byte-identical; the assertions got stronger, not looser.

Everything else is new tests, red as written.

Generated files

core/src/host/project-assistant.js and its test are mirrored whole, so this diff carries exactly
two generated paths: patches/ds-bot/files/src/host/project-assistant.js and
patches/ds-bot/files/test/project-assistant.test.js. No change to
patches/ds-bot/0002-flybot-api-host.patch and none to patches/ds-bot/files/src/client/*.

generate is not byte-idempotent in this repository: on this clean base it also rewrites
0001-core-fixes.patch, 0003-flybot-api-client.patch and 0014-flybot-branding.patch with
44 / 23 / 2 lines of pure hunk-header and count drift, and 0008-client-bundle.patch (192 content lines
differ in the patch, but the applied bytes are unchanged — check is green with the regenerated
0008 as well as with the base bytes). All four are restored to the base bytes, so the diff carries only the two mirrors; the
applied bytes are unchanged, and check stays green either way.

$ node scripts/ds-bot-patches.mjs check
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/

What I ran

At this tip, TZ=UTC, ARCH_DOCS_BASE=7f4aa2a70766a7292cf5474ec8a7bbc364f5d1ea:

• npm run verify — exit 0. Suites: core 2101, social 539, tree 125, timeline 129,
  workflow 289, .github 549, interop 37, live-e2e 24 — every suite fail 0.
• node core/scripts/build-client.mjs --check — silent, exit 0 (core/client.js untouched; the
  bundle does not carry project-assistant.js).
• node scripts/ds-bot-patches.mjs check — the three lines above.
• arch-docs: OK (5 changed file(s)) against the base ref above.
• focused file: 22/22 at TZ=UTC and 22/22 at TZ=Asia/Shanghai.

Not run: the real-model bench.mjs (minutes, spends tokens, not part of verify). No live instance
was touched.

Size

Measured as add + delete, generated files and mirrors excluded: 190 lines of change (+164/−26) —
source +48/−17 (65 changed lines), tests +101/−4, docs +15/−5.

Gates at this head

At f01a17c6c9a52c0e3e047447292960f22a12f9de, 13 of 14 check runs are green: tests-ok,
merge-gate, judge, package, agent-audience, reviewed tip (report only), risk tier (report only), plugins (social/tree/timeline/workflow), Node 24. enqueue-guard is red only
for the missing Verifier report:

do not enqueue #1761: no Verifier report names the tip f01a17c6c9a5

Its summary states the same fork rule the earlier PRs saw: the head lives in a fork: the issue reference and agent-pr rules are not asked of it. Nothing else refuses this head.

[HugoXK]

@Birfy please review this tip f01a17c6c9a52c0e3e047447292960f22a12f9de and run the gates on it.

State at this head (base 7f4aa2a70766a7292cf5474ec8a7bbc364f5d1ea, 5 files, +313/−47, Fixes #1758):

• checks: 13 of 14 green; the only red is enqueue-guard, whose whole reason is do not enqueue #1761: no Verifier report names the tip f01a17c6c9a5.
• the branch is rebased onto the live main and regenerated from zero there: the two patches/ds-bot/files/src/host/ mirrors carry the change, 0002/0008 are untouched, and the generate byte-drift on 0001/0003/0014 (pure hunk headers) was restored to its base bytes with the check re-run green afterwards.
• verify is exit 0 at this head: core 2101 / social 539 / tree 125 / timeline 129 / workflow 289 / .github 549 / interop 37 / live-e2e 24, zero failures; the new test file is 22/22 in both TZ=UTC and TZ=Asia/Shanghai.

[Birfy]

Verdict: PASS at f01a17c
Written-by: Reviewer
Attestation: f01a17c role=Reviewer

Ran at this tip:

$ cd core && TZ=UTC node --import ./test/harness-resolve.mjs --test test/project-assistant.test.js
# tests 22  # pass 22  # fail 0
$ cp /data/dsh/bot/workspace/1761base/core/src/host/project-assistant.js core/src/host/project-assistant.js && TZ=UTC node --import ./test/harness-resolve.mjs --test test/project-assistant.test.js
# tests 22  # pass 16  # fail 6
$ node scripts/ds-bot-patches.mjs check
ok   10 patches apply cleanly in order
ok   core/ equals ds-bot/ + patches/ds-bot/
$ node r1761-risk.mjs
risk-tier: T2; basis: core/src/host/project-assistant.js=T2, core/test/project-assistant.test.js=T1, patches/ds-bot/files/src/host/project-assistant.js=T1, patches/ds-bot/files/test/project-assistant.test.js=T1, docs/subsystems/core-host.md=T0

Review points:

• Red first, as required. With only core/src/host/project-assistant.js restored to the base 7f4aa2a7 blob and the tip tests kept, 6 of 22 fail — the same six the body names: 10 (held label), 12/13 (#P-194 due instant), 14 (#P-197 overflow), 16 (#P-195 waiting filter), 20 (card lists through the real task/board path). Restored, 22/22 green. The two changed existing tests are justified: test 10's old expectation asserted held from 09:00 to 08:45, a label that starts after it ends, and test 20's old fixture listed two open cards the new rule no longer calls waiting.
• The tree I ran is the tip, not a local rebuild: the git blob sha of all five changed files in the tarball of f01a17c6 equals the sha the PR files API reports for them (a532b032 source, 5ba8cca6 test, mirrors identical, ce840ca9 doc).
• Scope is exactly the five files the body names, +313/−47, one commit on 7f4aa2a7 — the live main today — so no stale-base risk. Both patches/ds-bot/files/ mirrors are byte-identical to their core/ twins; 0002/0008 untouched; check green without regenerating anything.
• The fix reads correctly: dayIn(at) defaults to DIGEST_ZONE (Asia/Shanghai), so Date.parse(day + 'T09:00:00+08:00') is the real due instant and clockIn(quiet.zone, …) compares it as a clock in the owner's zone; the held label now starts from held.at; the overflow line counts actionable.length - MAX_ACTIONABLE; actionableFor keeps a row whose status is empty rather than dropping it, and waitOf maps only blocked / result-ready to their own wording. withTitles resolves status from the task store, so the digest does not depend on the board carrying it.
• Docs move with the code (docs/subsystems/core-host.md, owned path) and describe the new label, the overflow line and the fork's exemption note; arch-docs is green in CI at this tip.
• Checks at this tip: 13 of 14 green — tests-ok, merge-gate, judge, package, the four plugin suites, Node 24, agent-audience, reviewed tip (report only), risk tier (report only). The only red is enqueue-guard, and its whole reason is no Verifier report names the tip f01a17c6c9a5 — this report answers it. Tier T2, so this PASS plus green CI is what the gate needs.
• Fork rule: the head lives in HugoXK/FlyBot, so the guard does not ask this PR for the checkpoint block, Owner:/Reviewer: lines, agent-pr or the issue reference. The body names Fixes #1758 anyway, and I did not edit the contributor's body to add metadata the fork exemption does not require.

Not run: the full core suite could not finish on this box — it died on pthread_create: Resource temporarily unavailable (7 forks, environment limit, not an assertion); CI tests-ok is green at this tip and covers it. No dev/bench.sh smoke: the change sits in a host module with no product-level path today (project-assistant-owner has no channel case), so the smoke bench would not exercise it; the body says the same and I found the endpoint gap at core/src/host/channel.js where it says.