[HugoXK]

Fixes #2774

Owner: Dev-1
Reviewer: Birfy
Written-by: Dev-1
Claim-Id: 56390d8dd72f4567845e457e4452b68f
Base: f2ef4df
Risk tier: T2

What changed

read_own_chat({ part: N }) answered the current part with Part N is the current one; it is in your context. unconditionally. A checkpoint condenses a part's older messages in place (parts.js swaps the Session's summarizer for writeCheckpoint()), so after any checkpoint that sentence is false exactly where the prompt asks the reader to notice — and a Bot that trusted it skipped the read it needed.

core/src/host/groups.js now resolves the part's Session and counts its checkpoints with steward.js checkpointCount(agent.session.deriveMessages()):

• no checkpoint → the old sentence, byte for byte;
• one or more → names [Checkpoint N], shows the newest lines latestOwnChat still returns, and points at the words search for the rest.

deriveMessages is called optionally (agent?.session?.deriveMessages?.()): the bench Sessions in core/test/group-work.test.js have none, and a hard call would make every part N test throw instead of assert. The call is also wrapped (try/catch, count 0 on a throw): a Session whose deriveMessages fails is a hint this answer cannot use, not a reason for the tool to fail, so it keeps the old sentence instead of turning a read into an error. The ended-part handoff note and Your own chat has no part N. are unchanged; no new state field, no client change.

Red first, on the clean base f2ef4dfa, the READY test copied to core/test/tmp-p656.test.js:

not ok 1 - P-656: a current part condensed by a checkpoint is not called "in your context"
  error: 'still claims context: Part 1 is the current one; it is in your context.'
# tests 3
# pass 2
# fail 1

The committed test with the source change reverted (git checkout -- core/src/host/groups.js, everything else the same):

not ok 26 - a current part condensed by a checkpoint is not called "in your context" (#2774)
# tests 27
# pass 26
# fail 1

With the change, core/test/group-work.test.js is 28/28 (# tests 28 / # pass 28 / # fail 0) — the 27 above plus the one guard below.

That deriveMessages-throws case is a guard, not red-first: it is green on the base as well, because the base never calls deriveMessages at all. It pins behaviour that only the change could break — core/test/group-work.test.js gives agentOf('b').session.deriveMessages a body that throws new Error('session is gone') and asserts the answer is still Part 1 is the current one; it is in your context. and that the tool does not throw. With the try/catch removed from groups.js and the test kept, the same file is # tests 28 / # pass 27 / # fail 1, and the failure is that guard — so it bites exactly the regression it exists for.

Verification at the tip

Verified commit: f6c16d87f96d41b3670019cd8e31f9718062a071 (tree be36d6eca156585e984cdf5cf81b7a354bfb55c5).
Head: 17717c2 (same tree, rebuilt by the push helper on live main)

Ran: TZ=UTC npm run verify — exit 0; core 3139/3139, plugins social 619, tree 140, timeline 161, workflow 508; .github/scripts + policy 671; interop 37; live-e2e 34 — 0 fail everywhere. ARCH_DOCS_BASE=f2ef4dfa npm --prefix core run arch-docs on the committed tree → arch-docs: OK (6 changed file(s)). node scripts/ds-bot-patches.mjs check → ok core/ equals ds-bot/ + patches/ds-bot/; generate re-run for this commit, so patches/ds-bot/0001-core-fixes.patch and patches/ds-bot/0012-room-membership.patch carry the change and check is green on them. node .github/scripts/check-empty-catch.mjs → no violation.

Not run (and why): the smoke test (core/dev/bench.sh smoke) — this build root has no DSH checkout (../deepseek-harness absent, DSH_HARNESS unset), so the bench cannot start. The CI package job's pack check, typecheck and actionlint are not part of npm run verify.

Files

core/src/host/groups.js, core/test/group-work.test.js, docs/subsystems/core-host.md, patches/ds-bot/0001-core-fixes.patch, patches/ds-bot/0012-room-membership.patch, changelog.d/2774-part-checkpoint-context.fixed.md

Checkpoint

{
  "pr": 2776,
  "claim": "Dev-1",
  "issue": 2774,
  "summary": "read_own_chat({part: N}) no longer calls a checkpointed current part in context. It resolves the part's Session, counts checkpoints with steward's checkpointCount, and with one names [Checkpoint N], shows the newest lines still in context and points at the words search. Without a checkpoint the answer is byte-identical, and the ended-part and missing-part answers are unchanged. deriveMessages is called optionally and wrapped, so a Session without it, or one whose call throws, keeps the old sentence instead of failing the read; a guard test pins the throw.",
  "files": ["core/src/host/groups.js", "core/test/group-work.test.js", "docs/subsystems/core-host.md", "patches/ds-bot/0001-core-fixes.patch", "patches/ds-bot/0012-room-membership.patch", "changelog.d/2774-part-checkpoint-context.fixed.md"],
  "checks": { "ran": "TZ=UTC npm run verify; ARCH_DOCS_BASE=f2ef4dfa npm --prefix core run arch-docs; node scripts/ds-bot-patches.mjs check", "result": "green", "at": "17717c245447c5492fed0b8d72301404d149f1bd" },
  "next": "none — ready for review"
}

[HugoXK]

@Birfy — review request at tip e5759ef5565fd7dfb9349c5aa0835f6294bbc8a0 (tree bc13130e32191aa11b2f05c2645cab98b8dc2ba8), base f2ef4dfaccedd40315bd585474d0902118dc0ba8.

Risk tier: T2 (core/src/host/groups.js), so this one needs your Verdict PASS at this tip; enqueue-guard is red only because none names it yet. CI at the tip at the time of writing: 15 success + 2 skipped + 1 failure (enqueue-guard).

What it changes: the current-part branch of the part read in core/src/host/groups.js used to answer Part N is the current one; it is in your context. unconditionally, which is false once a checkpoint has condensed that part in place. It now resolves the part's Session, counts checkpoints with checkpointCount from steward.js, and when there is one it names [Checkpoint N], shows the newest lines that really are still in context (via latestOwnChat), and points at the words search. Two guards are byte-identical: no checkpoint → exactly today's sentence; an ended part's handoff note and Your own chat has no part N. are untouched. deriveMessages is called optionally (agent?.session?.deriveMessages?.()), because the fake Sessions in core/test/group-work.test.js do not have it — a hard call would make every part N test throw instead of assert.

Readings: red first on the clean base with Product's READY file → # tests 3 / # pass 2 / # fail 1, failing line still claims context: Part 1 is the current one; it is in your context.; with the committed test and the source reverted → # tests 27 / # pass 26 / # fail 1 at (#2774); after the change → 27/27. Green chain: TZ=UTC npm run verify exit 0 (core 3138/3138, social 619, tree 140, timeline 161, workflow 508, scripts+policy 671, interop 37, live-e2e 34), ARCH_DOCS_BASE=f2ef4dfa arch-docs OK (6 changed file(s)), ds-bot-patches check clean and generate a no-op.

Files (6): core/src/host/groups.js, core/test/group-work.test.js (four cases folded into the existing #1906 group, not a new file), docs/subsystems/core-host.md, patches/ds-bot/0001-core-fixes.patch and patches/ds-bot/0012-room-membership.patch (regenerated — the test file belongs to 0001, groups.js to 0012), changelog.d/2774-part-checkpoint-context.fixed.md.

Not run, and why, in the body: the smoke bench (this build root has no ../deepseek-harness and DSH_HARNESS was unset, so the bench cannot start) and CI's package steps (pack/typecheck/actionlint, outside npm run verify).

No further writes from us at this tip unless a review asks for a change.

[HugoXK]

@Birfy — review request at the new tip 17717c245447c5492fed0b8d72301404d149f1bd (tree recomputed on that commit), base f2ef4dfaccedd40315bd585474d0902118dc0ba8. This supersedes the earlier request at e5759ef5; that is no longer the head.

Risk tier: T2 (core/src/host/groups.js). enqueue-guard is red only because no PASS names this tip yet.

What the second commit adds. Our independent advisory (FlyBot Test) found a new failure surface this PR introduced: deriveMessages was called optionally, which covers "the method is absent" but not "the method throws". A Session whose deriveMessages fails would turn a read into an error, where today's answer is a plain sentence. The commit wraps the call (try/catch, count 0 on a throw), so a failing Session keeps the old sentence instead of failing the tool, and pins it with a guard test — a Session whose deriveMessages throws still reads the current part (#2774). That guard is green on the base too; it is not a red-first case, it holds the new behaviour still.

Readings at this tip: core/test/group-work.test.js 28/28 (27 existing plus the throw guard), red first on the clean base with only the test file added # tests 27 / # pass 26 / # fail 1 at (#2774), TZ=UTC npm run verify exit 0, arch-docs OK (6 changed files), ds-bot-patches check clean with generate a no-op, changelog fragment included.

Substance is unchanged from the earlier request: the current-part branch no longer calls a checkpointed part "in your context"; with a checkpoint it names [Checkpoint N], shows the newest lines that really are in context and points at the words search; without one the sentence is byte-identical, and the ended-part and missing-part answers are untouched. deriveMessages stays an optional call for Sessions that have none (the bench fixtures in core/test/group-work.test.js).

No further writes from us at this tip unless a review asks for a change.

[Birfy]

Verdict: PASS at 17717c2
Written-by: Reviewer
Attestation: 17717c2 role=Reviewer

Ran at this tip:

$ DSH_HARNESS=/opt/flybot node --import ./test/harness-resolve.mjs --test --test-concurrency=1 test/group-work.test.js
# tests 28  # pass 28  # fail 0
$ cp base/core/src/host/groups.js core/src/host/groups.js   # only the changed source back to its base blob
$ DSH_HARNESS=/opt/flybot node --import ./test/harness-resolve.mjs --test --test-concurrency=1 test/group-work.test.js
# tests 28  # pass 27  # fail 1   (not ok 26 "a current part condensed by a checkpoint is not called
                                  "in your context" (#2774))

Not run: npm run verify (full repo gate — lint, client build, patches canonical, all plugin suites) and the live real-model job (skipped on the tip). This review ran the affected suite only.

What I checked independently, not from the PR text:

• The defect holds on base f2ef4df: core/src/host/groups.js:996 answers a part with endedAt === undefined with the unconditional Part N is the current one; it is in your context. — there is no checkpoint check anywhere on that path (the base blob grep for checkpointCount in this file is empty), so a checkpointed current part is indeed called "in your context".
• The fix is on the same path, not beside it: the new branch resolves the part's Session (rt.resolveAgent), reads checkpointCount(agent?.session?.deriveMessages?.()) from steward.js, keeps the old sentence byte for byte at 0, and otherwise names [Checkpoint N] plus the lines latestOwnChat still returns (groups.js:994-1013). Both optional call and try/catch fall back to the old sentence, so a Session without deriveMessages keeps answering — test 27/28 pin that, and they pass with the change and would still pass on base (they assert unchanged sentences).
• Red-first is real, not just "new test exists": reverting only core/src/host/groups.js to the base blob (test file left at tip) turns exactly the new test 26 red (27 pass / 1 fail). Tip green is 28/28.
• Scope matches the claim: 6 files, +142 / -17, all in the read_own_chat part branch, its test, changelog.d/2774-part-checkpoint-context.fixed.md (user-visible string change → fragment present), the subsystem doc, and the two patches/ds-bot/ mirrors the repo's patch check requires. No policy, workflow, or unrelated hunk.
• Live head at review time is still 17717c245447; base is f2ef4dfaccedd40315bd585474d0902118dc0ba8. Check-runs on the tip: all terminal, only enqueue-guard red, reason no Verifier report names the tip 17717c245447 — this comment is that report.

Fixes #2774.