GH Issues Orchestrator automates the lifecycle of GitHub issues: it fetches issues from a repository, presents them for confirmation, runs pre-flight checks, and spawns focused sub-agents (workers) to implement fixes, push branches, and open pull requests. It also monitors PRs for review comments and can spawn review-fix agents to address requested changes. The skill relies on curl + the GitHub REST API (no gh CLI) and uses token-based auth (GH_TOKEN). It supports interactive, watch, and cron modes, forked workflows (push to a fork, PR to upstream), claim-based concurrency protection, and configurable polling.
Use when you want to automate triage and remediation of small-to-medium GitHub issues, run periodic cron-driven repairs, or continuously monitor PRs for actionable review comments. Ideal for repositories where repeatable, code-fixable issues appear and you want automated agents to open focused PRs. Not suitable for large-scope architectural changes that require human design.
Designed for OpenClaw-style agent runtimes that can exec shell, run curl/git, and spawn sessions (e.g., subagents/sessions_spawn).
gh-issues is a comprehensive GitHub issue orchestration skill that fetches issues, spawns sub-agents to implement fixes, and monitors PR review comments. It uses curl + GitHub REST API exclusively (no gh CLI). The skill_md is thorough with 6 well-defined phases, argument parsing, fork mode, cron mode, watch mode, and claim-based deduplication. No bundled scripts to test.
Security deductions: -12 for shell injection risk (unquoted variables in curl commands like {SOURCE_REPO}, {title}, {body} in JSON templates), -15 for credential exposure (token in git remote URLs visible via git remote -v), -8 for global git config modification (credential.helper cleared globally), -3 for jq string interpolation without sanitization. The skill is well-structured and genuinely useful but the security model relies on the agent executing commands in a controlled environment. No evidence of malicious intent. No curl|bash, no hardcoded credentials, no data exfiltration, no auto-update mechanisms. The skill uses proper GitHub API auth patterns. Architecture is strong with clear phase separation, though the monolithic SKILL.md could benefit from splitting scripts/ references.
