gh-issues

• You have a backlog of clearly-scoped bug issues and want a first-pass automated fix attempt before human review.
• You run a scheduled CI/cron pipeline and want issues automatically triaged and PRed without manual intervention (--cron mode).
• You work with a fork-based contribution model and need branches pushed to your fork with PRs targeting an upstream repo (--fork).
• You want PR review comments addressed automatically — use --reviews-only to have the skill scan open fix/issue-* PRs and push fixes for requested changes.

• Issues are vague, architectural, or large in scope — the skill's sub-agents self-assess confidence and will skip issues they rate below 7/10, but highly complex issues should be handled manually.
• You need guaranteed correctness — sub-agents run tests when available, but auto-generated fixes should be reviewed before merging into production.
• You have no GitHub token (GH_TOKEN) — the skill cannot authenticate to any GitHub API without it.

Up to 8 sub-agents run concurrently, each tackling a single issue end-to-end: branch creation, code analysis, fix implementation, test execution, commit, push, and PR opening. Each agent has a 60-minute budget and a built-in confidence check — it skips issues it cannot address with ≥7/10 confidence.

After PRs are opened, Phase 6 fetches reviews, inline comments, general PR comments, and even embedded bot reviews (e.g., Greptile). Actionable comments trigger additional sub-agents that check out the branch, apply the requested changes, push an update, and reply to each comment with the commit SHA.

Pass --fork user/repo to push all branches to your fork and open PRs from the fork targeting the upstream source repository — matching the standard open-source contribution workflow without requiring write access to the upstream.

--cron enables fire-and-forget scheduling: the orchestrator picks one eligible issue, spawns a sub-agent, and exits immediately — safe to call from a cron job every few minutes. --watch keeps the orchestrator alive, polling for new issues and review comments on a configurable interval.

Before spawning, the skill checks for existing open PRs, in-progress branches on the push repo, and a local claims file with 2-hour expiry. This prevents duplicate sub-agents from racing on the same issue across overlapping cron runs or watch cycles.

Pass --notify-channel with a Telegram channel ID to receive a final summary message containing PR links and per-issue results. Only the final outcome is sent — not intermediate status updates.

Run /gh-issues owner/repo --label bug --limit 10 --cron from a scheduled job. Each invocation picks the next unclaimed bug issue, spawns a sub-agent to implement a fix, and exits. Over time, the backlog is worked through incrementally without human involvement between runs.

Use /gh-issues upstream/project --fork myfork/project --label good-first-issue to have the skill fetch issues from the upstream repo, push fix branches to your fork, and open PRs targeting upstream — respecting the standard contributor workflow.

Schedule /gh-issues owner/repo --reviews-only --cron alongside your existing PR pipeline. Each run scans open fix/issue-* PRs for unaddressed CHANGES_REQUESTED reviews or inline comments, picks the first unhandled PR, and spawns an agent to apply the requested changes and reply.

Run /gh-issues owner/repo --label enhancement --watch --interval 10 during a sprint planning session. The skill fetches issues, asks you to confirm which to process, spawns agents, and polls every 10 minutes for new issues and review feedback — presenting cumulative results until you say 'stop'.