gh-issues_skill

Overview

This skill automates fetching GitHub issues, spawning parallel sub-agents to implement fixes, opening pull requests, and monitoring review comments until they are addressed. It works without the gh CLI, using curl and the GitHub REST API with a provided GH_TOKEN. The skill supports fork mode, watch/cron modes, dry-run, and notification via a Telegram channel.

How this skill works

It runs a strict 6-phase flow: parse arguments, resolve tokens and fetch issues, present and confirm which issues to handle, run local git pre-flight checks, spawn sub-agents to implement fixes and open PRs, then monitor PR review comments and optionally reply. Sub-agents are invoked via sessions_spawn with configurable concurrency, model selection, and claim tracking to avoid duplicate work. Cron mode supports fire-and-forget single-issue processing with a cursor and claims file for safe parallelism.

When to use it

• Automatically triage and fix a batch of issues filtered by label, milestone, assignee, or state.
• Run on a schedule (cron) to continuously process backlog items without manual intervention.
• Watch mode to poll a repo and auto-process newly created issues over time.
• Only address PR review comments (reviews-only) to keep PRs moving toward merge.
• Dry-run to inspect what would be processed before launching sub-agents.

Best practices

• Ensure GH_TOKEN is set or configured in the OpenClaw manifests before running to avoid auth failures.
• Use fork mode when you want fixes pushed from your fork; verify the fork remote is added or allow the skill to add it.
• Start with conservative limits (--limit, --interval) and enable --yes for unattended runs when confident.
• Monitor the claims and cursor files under /data/.clawdbot to understand in-progress work and avoid collisions.
• Use --notify-channel to receive a concise final PR summary instead of noisy intermediate updates.

Example use cases

• /gh-issues openclaw/openclaw --label bug --limit 5 --assignee @me — fetch top 5 bug issues assigned to you and spawn fix agents.
• Run as a cron job with --cron and --fork user/repo to sequentially process backlog without waiting for results.
• Use --reviews-only to scan open PRs in the repo and spawn agents to address review comments.
• Run with --dry-run to preview matching issues and confirm filters before committing to automation.
• /gh-issues owner/repo --milestone v1.0 --model glm-5 --notify-channel -1002381931352 to target a release milestone with a specific model and final notifications.