The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/close_issue.py uses subprocess.run to invoke the GitHub CLI (gh) for issue management tasks like editing bodies, closing issues, and updating milestones.
Evidence: The gh and graphql functions in the Python script construct command lists that are executed without a shell, preventing shell-based command injection.
[INDIRECT_PROMPT_INJECTION]: The skill reads and parses GitHub issue bodies, which are external data sources that could potentially contain malicious instructions.
Ingestion points: The read_issue function fetches issue text using the GitHub GraphQL API.
Boundary markers: The script identifies logical sections using regex headers (e.g., ## Reproduction) but does not use specialized security boundaries for the content.
Capability inventory: The skill can modify repository content and state through issue edits and closures.
Sanitization: Ingested content is parsed for status information (checkboxes, sub-issues) but not explicitly sanitized for prompt-based attacks; however, the script's logic primarily enforces a restrictive verification workflow.
[EXTERNAL_DOWNLOADS]: The skill retrieves metadata and content from GitHub repositories.
Evidence: Authenticated API requests to GitHub are used to gather data necessary for the verification logic. GitHub is a well-known and trusted service for development workflows.

close-issues