The Agent Skills Directory

[COMMAND_EXECUTION]: The main script (audit_issues.py) uses subprocess.run to invoke the GitHub CLI (gh) for performing repository operations like querying the GraphQL API and creating or editing issues. This is a standard integration method for repository management skills.
[DYNAMIC_EXECUTION]: The auditing process establishes a dependency on the close-issues skill, executing its close_issue.py script via sys.executable during the apply phase. This cross-skill interaction assumes the local availability and integrity of the referenced script.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from GitHub issues, creating a potential attack surface.
Ingestion points: Issue titles and bodies are fetched from remote GitHub repositories in audit_issues.py via the synced_issues and read_issue functions.
Boundary markers: There are no explicit delimiters or safety instructions used when the agent is prompted to read and summarize issue content.
Capability inventory: The script possesses capabilities to create, edit, and close issues, as well as write local files for caching and reporting.
Sanitization: Untrusted content is used for signal extraction via regular expressions and is written to audit plans without specific sanitization against instruction injection.
[SAFE]: The skill implements a robust 'plan-then-apply' workflow. It generates a markdown summary (page) for the user to review and provides an 'approval code'—a cryptographic hash of the plan and verdict files—to ensure the audit plan has not been tampered with or modified between the approval and execution steps.

audit-issues