Issue Context

When you assign the bot to an issue or ask it to address review feedback on a PR, it fetches the issue or PR title, body, and visible comments and includes them in the prompt sent to the coding agent. The bot also pulls in related issue context — useful for sub-issues that depend on a parent for their rule or success criteria.

What gets pulled in

The bot collects related issues from two sources, in this order:

Native parent issue. When the target is an issue, the bot queries GitHub’s Issue.parent field. If the issue was added as a sub-issue of another via the GitHub UI or API, that parent is fetched and labeled as the parent in the prompt. PRs do not have a parent field, so this step is skipped for PRs.
Textual references in body and comments. Two reference shapes are recognized:
- Short form: #NNN — always treated as same-repo.
- Full URL form: https://github.com/<owner>/<repo>/issues/NNN — only followed when <owner>/<repo> exactly matches the current repo (case-insensitive).

The native parent (when present) is always fetched first and labeled Parent issue. Textual references are labeled Reference. If the body also mentions the native parent textually, the duplicate is dropped so the parent is only fetched once.

Hiding comments from context

If you minimize a comment on an issue or PR using GitHub’s Hide option, the bot skips it when collecting context. Use this to keep noise out of the prompt (off-topic chatter, resolved tangents, outdated suggestions) without deleting the comment.

This applies to both the target issue or PR and to any linked issues fetched as related context.

Safety rules

Same-repo only. Cross-repo URL references are skipped. Short-form #NNN is resolved against the current repo only.
One hop only. The bot does not recurse into the parent’s parent or references found inside linked issues.
Breadth cap. At most 5 linked issues are fetched per task. The parent (when present) takes the first slot, leaving 4 for textual references. Extras are dropped in the order they appear.
Self-reference dedup. If the target issue references itself or claims itself as its parent, the self-reference is dropped.
Tolerate missing refs. If a linked issue can’t be fetched (404, 403, etc.), the bot logs and skips it. The task continues with whatever links did resolve.
No PR URLs. Only /issues/NNN URLs are matched. /pull/NNN URLs are ignored.

Where it applies

Linked-issue context is included in three places:

When the bot is asked to analyze an issue or PR.
When the bot opens a PR for an assigned issue.
When the bot addresses review feedback on a PR — the PR body’s Closes #N pulls in the parent issue’s rule automatically.

Example

Issue #215 is a sub-issue of #111 and its body says:

Verify each method against the rule in #111.
Part of #111.

When the bot is assigned to #215, it queries Issue.parent, finds #111, and fetches it. The textual #111 mentions are deduped against the parent so it is only fetched once. The prompt the coding agent receives includes the full body and comments of #111 as context.

If #215 had no textual reference to #111 and was only linked via the native sub-issue relationship, the same parent context would still be pulled in.