Skip to content

0.31.2–0.31.4: bug-hunt batches B (recall), D (dashboard) and E (housekeeping) - #82

Merged
Interstellar-code merged 28 commits into
mainfrom
claude/wonderful-bell-5nyrqz
Nov 8, 2026
Merged

0.31.2–0.31.4: bug-hunt batches B (recall), D (dashboard) and E (housekeeping)#82
Interstellar-code merged 28 commits into
mainfrom
claude/wonderful-bell-5nyrqz

Conversation

@Interstellar-code

@Interstellar-code Interstellar-code commented Nov 8, 2026

Copy link
Copy Markdown
Owner

Brings in PRs 78, 79 and 80 with their own commits, so GitHub shows them as merged, plus the release bookkeeping and one harness test fix. Each batch is its own release in the CHANGELOG.

Fixes #19
Fixes #7
Fixes #12
Fixes #17
Fixes #26
Fixes #9
Fixes #23
Fixes #20
Fixes #22
Fixes #29
Fixes #30
Fixes #31
Fixes #32
Fixes #33
Fixes #36
Fixes #37
Fixes #38
Fixes #39

Releases

Harness test fix (05b10ee)

#38 releases a marker whose model call failed with release <root> <id> --failed. tests/session.test.ts's releasesOf read the last argument as the id, so it got --failed (SESSION.1.a4). It now takes the last non-flag argument, and the spec checks that only the lost call is released as failed. The batch session could not run the claude plugin test harness.

Decisions accepted (details in each PR)

These are conservative choices: #19 does a linear cosine scan over embedded rows, #17 no longer fires on recently, and #38 backs off 1h, doubling to a 24h cap, giving up after 5 failed model calls with an activity line.

Tests

  • claude plugin test .: 544 pass, 0 fail.
  • Full bun test ./tests/bun ./engine/test: 1778 pass, 2 skip, 9 fail. Those are the same 9 as on main in this container.
  • engine-diff ok (build .18).

Generated by Claude Code

claude added 28 commits November 8, 2026 14:47
@claude
Skip session-pending markers that vanish before they are stat-ed (#36)
146f132 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01M2ePSnY9UQPcd6a4DCMpXD
@claude
Put back a fresh sweep lock found during stale takeover instead of de…
292839e 
…leting it (#37)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01M2ePSnY9UQPcd6a4DCMpXD
@claude
Back off and give up on session summaries whose model call keeps fail…
9b595c6 
…ing (#38)

A failed or unanswered summary model call now releases the marker with
attempts+1 and notBefore = now + 1h doubling per attempt (cap 24h);
session-pending list skips markers still waiting, and the fifth failed
attempt drops the marker with one activity line. Old markers parse as
attempts 0; session-pending write starts again at 0.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01M2ePSnY9UQPcd6a4DCMpXD
@claude
docs(plans): Status lines say what shipped; the sweep marker is writt…
8f2a6a9 
…en first by design (#39)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01M2ePSnY9UQPcd6a4DCMpXD
@claude
fix(engine): resolve 'day before yesterday' to two days back (#9)
07fadc3 
The yesterday check ran first and matched the phrase; check the two-day form first.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01M2ePSpuvDs1ukDqG5dKSZu
@claude
Match confidence boosters as whole words, not substrings (#23)
31fd224 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01M2ePSpuvDs1ukDqG5dKSZu
@claude
fix(engine): match temporal trigger keywords on word boundaries (#17)
d84b59f 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01M2ePSpuvDs1ukDqG5dKSZu
@claude
Forward useSynonyms, useIntent, useMmr and mmrLambda through the memo…
0220502 
…ry facade (#7)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01M2ePSpuvDs1ukDqG5dKSZu
@claude
fix(recall): vector similarity covers every visible embedded row, not…
0d282f8 
… the 500 newest (#19)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01M2ePSpuvDs1ukDqG5dKSZu
@claude
Keep every query word in the FTS query; synonyms fill the remaining r…
0b751e7 
…oom (#26)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01M2ePSpuvDs1ukDqG5dKSZu
@claude
fix(engine): expand words in several synonym groups to every group (#12)
8194264 
getSynonyms now returns the canonicals and synonyms of every group a word
belongs to, deduped in group order; normalizeQuery keeps a single canonical.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01M2ePSpuvDs1ukDqG5dKSZu
@claude
fix(engine): every query word and every synonym group fits inside the…
b3a159c 
… recall token cap (#12)

A word in several synonym groups (push: deploy + git) now expands to 12+
tokens, so a flat expansion let it crowd later query words and its own
second meaning out of the 12-term FTS cap and the 6-term fact fallback.
expandedTokens now lists the literal query words first and then takes one
variant from each synonym group of each word in turn (new getSynonymGroups
in synonyms.ts). The token set is unchanged; only the order differs. The
FTS cap stays 12, as decided for #26.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01M2ePSpuvDs1ukDqG5dKSZu
@claude
Session markers: a notBefore past the longest back-off is not trusted…
ed2c4a2 
…; comments name the back-off (#38)

Follow-up to the review of the #38 fix: a marker whose notBefore is more
than 24 hours ahead (a clock that jumped back after a failed attempt, or a
hand edit) was hidden from listSessionPending for as long as it said, so it
was never retried or dropped. Such a wait is now treated as due. The
finishSessionSummaries comment and the bin/memories.ts usage line describe
the back-off and release --failed.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01M2ePSnY9UQPcd6a4DCMpXD
@claude
Memories list: only the reply to the latest load writes the rows (#20)
d701544 
A filter change that overlapped an earlier /api/memories request could
let the older reply land last and overwrite the table with the older
filter's rows; pickTagFromSheet always raced (untagged first-open load,
then the tagged one). loadMemories now numbers each request and drops
stale replies and stale errors silently.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01M2ePSrn4zdhsS2cocxtA3V
@claude
Run a search typed after Escape while a request is in flight (#22)
7d595c4 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01M2ePSrn4zdhsS2cocxtA3V
@claude
Keep the knowledge note's Clear button across re-renders (#29)
6d8f220 
renderKnowledgeControls rebuilt the note on every render, replacing the
Clear button and dropping keyboard focus on each live event. Rebuild only
when the note's text or Clear presence changes.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01M2ePSrn4zdhsS2cocxtA3V
@claude
Show the error note when a review-queue reload fails over an earlier …
4b3e564 
…queue (#30)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01M2ePSrn4zdhsS2cocxtA3V
@claude
Keep the shown search hits after a forget instead of collapsing to 10 (…
d9dbcf1 
…#31)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01M2ePSrn4zdhsS2cocxtA3V
@claude
Keep the shown count after a forget on the first page too, without ba…
fef7919 
…ckfilling to 10 (#31)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01M2ePSrn4zdhsS2cocxtA3V
@claude
Live recall sets the star's last-recalled time (#32)
b6efa14 
recallMemory now sets star.lastRecalled to the live event's timestamp
(carried on the spark) or to now, beside the recall count.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01M2ePSrn4zdhsS2cocxtA3V
@claude
Settings: only the last copied command's button says Copied (#33)
3715727 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01M2ePSrn4zdhsS2cocxtA3V
@claude
Settings: let go of copy buttons from earlier draws (#33)
80b693f 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01M2ePSrn4zdhsS2cocxtA3V
@claude
Merge PR #78 (bug-hunt batch B: recall and query)
a8db8f7 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01M2ePSobwTXWVRg7fVZcKYo
@claude
chore(release): 0.31.2, engine build .18 with changes 34-40 (#19, #7, #…
7fb7348 
…12, #17, #26, #9, #23)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01M2ePSobwTXWVRg7fVZcKYo
@claude
Merge PR #79 (bug-hunt batch D: dashboard)
cdfdbd3 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01M2ePSobwTXWVRg7fVZcKYo
@claude
Merge PR #80 (bug-hunt batch E: housekeeping)
663b6e0 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01M2ePSobwTXWVRg7fVZcKYo
@claude
chore(release): 0.31.3 and 0.31.4 changelog (#20, #22, #29, #30, #31, #…
a09ba22 
…32, #33; #36, #37, #38, #39)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01M2ePSobwTXWVRg7fVZcKYo
@claude
test(session): the harness rig reads the session id past #38's --fail…
05b10ee 
…ed flag (#38)

tests/session.test.ts releasesOf took the last argument as the session id;
a failed model call now releases with 'release <root> <id> --failed', so it
read '--failed'. It now takes the last non-flag argument, and SESSION.1.a4
checks that only the lost model call is released as failed. claude plugin
test: 544 pass, 0 fail.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01M2ePSobwTXWVRg7fVZcKYo
@Interstellar-code
Interstellar-code merged commit fe1f2c6 into main Nov 8, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment