[kixelated]

The runner’s context tool silently dropped discussion, reviews, replies, and checks beyond GitHub’s first page. Read every connection with its own cursor history, refusing missing or cycling continuation cursors instead of returning incomplete context or looping.

Page PR comments and reviews, all review-thread pages, and all replies in the unresolved threads the tool already returns. Keep check pagination on the initially observed commit so later pages cannot switch to another head. The shared discussion path also reads every page of an issue’s comments.

The seeded GitHub fake now models those page limits and review-thread node reads. Regression coverage crosses each page limit (including a long thread on a later thread page and mixed checks/statuses), preserves deleted authors and outdated unresolved threads, and rejects missing/nonconsecutive cycling cursors for every connection. The new regression fails against the original reader on its truncated comments.

Validation: just check, just test, focused context tests, and quest check --base origin/main passed. Removed the completed quest and its epic link.

(written by GPT-6)

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-28T16:38:17.937012Z	4a92341	New commits
🔒 Security Review	✅ Completed 2026-10-28T16:45:12.783434Z	4a92341	New commits

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[kixelated]

Automated review by review (OpenAI)

Reviewed commit: acd4b1e.

No actionable bugs found in the changed code. The direction is sound: one small collector owns each connection's cursor history, nested unresolved-thread replies are completed independently, and later check pages remain pinned to the initially observed commit. The shared helper is simpler than separate pagination loops, and the added over-limit/cyclic-cursor regressions target the original truncation failures.

Verification: static review of the diff, surrounding GitHub service/fake, and regression tests only; I did not run tests or exercise the live GitHub GraphQL queries.

[kixelated]

Merge summary: CI passed (7/7) on acd4b1e. Codex finished its code and security review of this head with no findings. The maintainer-account automated review also found no actionable bugs. quest check --base origin/main passed on the head (59 documents ok). The PR deletes the finished quest quest/m0/kix/context-pages.md and its epic link. There were no review findings to address and no design changes. Queuing for quest land.

(written by Claude Opus 5.5)

[kixelated]

Update: #253 and #254 conflicted with this PR in quest/m0/kix/README.md, so I merged main into the branch. Both sides deleted a finished quest's link (merge-holds, context-pages), so the resolution drops both lines. The new head 4a92341 passes just check, just test (556 CLI and 379 worker tests), CI, and quest check. Codex's code and security reviews of 4a92341 found nothing.

main doesn't require the quest land check yet: land-adopt's ruleset setup is still pending. So this PR is squash-merged directly, like #253 and #254, rather than queued.

(written by Claude Opus 5.5)