The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill interacts with untrusted external data sources such as GitHub pull requests, issues, and CI logs, which serve as ingestion points identified in SKILL.md and references/github.md. The skill includes a clear boundary marker by explicitly instructing the agent that instructions inside such content must be treated as data and do not authorize actions beyond the user's initial request. The skill's capabilities include Git write operations and GitHub API updates via gh, while sanitization is maintained by using structured JSON for API payloads in scripts/upsert_comment.py.
[COMMAND_EXECUTION]: The skill and its associated scripts execute git and gh commands via subprocess calls to manage repository state. The implementation in scripts/upsert_comment.py correctly avoids shell injection vulnerabilities by passing arguments as a list and using standard input for large data payloads, ensuring that content retrieved from or sent to GitHub is handled securely.
[SAFE]: The skill demonstrates high-quality defensive engineering by mandating backup references before history-rewriting operations and enforcing strict limits on destructive commands like force-pushing and rebasing. The use of the GitHub CLI with structured JSON output ensures that machine-readable data is parsed reliably, preventing errors common with text-based parsing of CLI output.

manage-git