[rtibbles]

Summary

• Pull requests to a non-default branch never link the issue their closing keyword names, so they are labelled unlinked and closed.
• Link those keyword references the way GitHub does on the default branch.
• When nothing is linked, link the one mentioned issue assigned to the author, ask which when several are, and tell an unlinked author the issue comes first.

References

Closes #112. Stacked on: #136. Unlinked keyword: learningequality/kolibri#15763.
Mutation run with the LE bot app token: learningequality/test-actions#104.

Reviewer guidance

JS tests pass on CI.

• issues: write and pull-requests: write on the app token are enough for addCloseIssueReferences: Test: addCloseIssueReferences with the bot app token test-actions#104 was linked to Test: link this issue to a pull request with the bot app token test-actions#103 on a non-default base.
• Off the default branch, a keyword reference to someone else's issue now closes the pull request, as it would on the default branch.

AI usage

Used Claude Code to revise the issue and implement the parsing, linking and replies. Verified with the full test suite, ESLint, Prettier, and the mutation run in test-actions.

🤖 Generated with Claude Code

https://claude.ai/code/session_013PjXpbp1QSbkpzvnRwm7yv

[rtibblesbot]

🟡 Waiting for changes

Outcome: reviewed 9614aa9

Last updated: 2026-11-10 08:47 UTC

[rtibblesbot]

PR #138: one suggestion inline. Linking flow matches #112. Reply tests cover each case through a fake github.graphql. CI passing.

• suggestion: a #N above GraphQL's 32-bit Int fails the whole reference query (inline).

Next step: Needs a full code review by a human engineer.

• Workflow steps changed: .github/workflows/contributor-pr-reply.yml.
• Confirm an assigned issue the author only mentions ("part of #7") should get a closing link, so that merging closes it.
• Confirm the app token's issues: write permission allows addCloseIssueReferences, since the PR says this has not been tested.

---

Written by rtibblesbot, an LLM-based coding agent.

[rtibblesbot]

PR #138: a single assigned candidate is still never linked, contrary to #112 (inline).

1 of 2 prior findings resolved. CI passing. Commits pushed after the review passes were not reviewed. The open finding is confirmed at fdf872e.

• blocking: a single assigned candidate is never linked, contrary to #112 (inline).

Prior-finding status

UNADDRESSED — scripts/contributor-pr-reply.js:82 — single assigned candidate is never linked
RESOLVED — scripts/utils.js:327 — #N above 32-bit Int fails the whole reference query

Next step: Needs a full code review by a human engineer.

• Workflow steps changed: .github/workflows/contributor-pr-reply.yml.
• Decide whether a mention-only reference to one assigned issue is auto-linked (as #112 specifies) or only asked about (as this PR does).
• Check that auto-linking via addClosingIssueReference cannot close the wrong issue when the PR body mentions an issue in passing.

---

Written by rtibblesbot, an LLM-based coding agent.

[rtibblesbot]

PR #138: both prior findings are resolved. No new findings. CI is passing.

Prior-finding status

RESOLVED — scripts/utils.js:327 — #N above 2147483647 fails the whole reference query
RESOLVED — scripts/pull-request-link.js:23 — one open candidate assigned to the author is never linked

Next step: Needs a full code review by a human engineer.

• Workflow steps changed: .github/workflows/contributor-pr-reply.yml.
• Check that linking an issue named in the PR body is what #112 intended, for an issue assigned to the author but not referenced with a closing keyword.
• Check the reply that asks contributors which issue they mean when several are assigned to them; its wording will be copied across repositories.

---

Written by rtibblesbot, an LLM-based coding agent.