The Agent Skills Directory

[SAFE]: The skill is primarily documentation-based, providing references for CI commands and safety protocols. The manifest.json and SKILL.md frontmatter confirm that no tools are provided, meaning the agent has no direct execution capabilities through this specific skill.
[COMMAND_EXECUTION]: The file scripts/_skill_cli.py contains a run_command function using subprocess.run. This is a shared helper script that follows security best practices by passing commands as lists rather than shell strings. It is not exposed as an active tool in the skill manifest.
[INDIRECT_PROMPT_INJECTION]: The skill defines patterns for processing external data such as GitHub Pull Request comments.
Ingestion points: External data enters through PR comments and interactive Q&A commands like /heyeddi ask (documented in reference/commands.md).
Boundary markers: The skill enforces a strict 'authorize merge' confirmation phrase required in the current conversation turn to prevent unauthorized actions.
Capability inventory: No subprocess or network tools are registered in the manifest.json for this skill.
Sanitization: The skill relies on behavioral instructions and anti-patterns to ensure the agent does not interpret data as commands without proper authorization.

heyeddi-ci-guide