The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides commands that facilitate the ingestion of untrusted external data from GitHub into the agent's processing context.\n
Ingestion points: Commands like gh pr view, gh issue view, and gh api (detailed in references/pr.md, references/issues.md, and references/api.md) are designed to retrieve potentially malicious content from pull requests, issues, and API responses.\n
Boundary markers: The skill does not provide instructions for wrapping external content in delimiters or include warnings to disregard embedded instructions within the fetched data.\n
Capability inventory: The skill utilizes the Bash tool (as defined in SKILL.md) to execute GitHub CLI commands, which provides the agent with a wide range of operational capabilities.\n
Sanitization: There is no evidence of content sanitization, escaping, or validation performed on the external data within the provided command examples.\n- [SAFE]: The skill exclusively uses the official GitHub CLI (gh) for all operations, ensuring that repository interactions follow established and secure protocols.\n- [SAFE]: No malicious patterns, such as obfuscation, hardcoded credentials, or unauthorized network exfiltration, were detected in any of the skill files.\n- [SAFE]: The instructions encourage secure practices by directing users to the official gh auth command set for authentication rather than handling raw secrets or tokens.

github