gh-issues

The skill operates as an orchestrator across six sequential phases:Parse Arguments — Reads flags like --label, --limit, --fork, --watch, and --model to configure behavior. Auto-detects the repository from the current git remote if not specified.Fetch Issues — Queries the GitHub Issues API using your token, applying label, milestone, assignee, and state filters. Pulls requests are automatically excluded from results.Present & Confirm — Displays a table of matching issues and asks which to process (or auto-processes all when --yes is set). Dry-run mode stops here for safe previewing.Pre-flight Checks — Validates git state, remote access, token validity, and existing PRs or in-progress branches to avoid duplicate work. Uses a claims file to coordinate between concurrent cron runs.Spawn Sub-agents (Parallel) — Launches up to 8 concurrent fix agents, each responsible for one issue: checking out a branch, analyzing the codebase, implementing the fix, running tests, committing, pushing, and opening a PR.PR Review Handler — Scans open fix/issue-* PRs for actionable review comments (including inline comments, general PR comments, and embedded tool feedback like Greptile). Spawns follow-up agents to implement changes and reply to reviewers.In cron mode, the orchestrator processes one issue per invocation and exits immediately after spawning — safe for scheduled task runners. In watch mode, it polls continuously at a configurable interval, tracking processed issues and addressed comments across cycles.

Parallel Sub-agents — Up to 8 concurrent fix agents work simultaneously, each handling one issue end-to-end within a 60-minute budget.Fork Mode — Push branches to your fork and open PRs targeting the upstream repo, enabling contribution to repositories you don't have write access to.PR Review Handling — Automatically detects actionable review comments (including CHANGES_REQUESTED reviews, inline code comments, and embedded tool feedback), implements changes, and replies to each comment with a commit reference.Confidence Gating — Each sub-agent self-assesses confidence (1–10) before implementing a fix. Issues scoring below 7 are skipped with a detailed reason rather than producing a poor-quality PR.Duplicate Prevention — Pre-flight checks detect existing PRs, in-progress branches, and claimed issues (via a shared claims file) to prevent redundant work across parallel or scheduled runs.Cron & Watch Modes — Cron mode processes one issue per invocation and exits immediately; watch mode polls continuously, auto-processing new issues and reviews on each cycle.Flexible Filtering — Filter issues by label, milestone title, assignee (including @me), state, and count limit before any processing begins.Telegram Notifications — Optionally send PR summaries and review completion notices to a Telegram channel via the --notify-channel flag.Model Selection — Override the default sub-agent model with --model to use any supported model (e.g., glm-5).

GitHub Token (required) — A GitHub Personal Access Token (classic or fine-grained) with repo scope for reading issues, creating branches, pushing code, and opening pull requests. Set as GH_TOKEN in your environment or configure it in the OpenClaw dashboard under the gh-issues skill entry. Supports: GitHub.com, GitHub Enterprise (via API base URL).Telegram Bot Token (optional) — Required only if using --notify-channel for PR and review completion notifications. Configure via the OpenClaw messaging integration. Supports: Telegram.

Bug Backlog Reduction — Run /gh-issues owner/repo --label bug --limit 10 to automatically triage and fix your top 10 open bug reports overnight, waking up to open PRs ready for human review.Scheduled Maintenance — Use --cron with a task scheduler to process one issue per hour continuously, keeping a steady flow of automated fixes without overwhelming reviewers.Open Source Contribution — Use --fork your-username/repo to contribute fixes to repositories you don't own, automatically opening PRs from your fork to the upstream project.PR Review Automation — Run with --reviews-only after a review cycle to have the skill address all pending CHANGES_REQUESTED comments and inline feedback across your open PRs in one pass.