The Agent Skills Directory

[SAFE]: The skill is primarily documentation-based, providing a reference for HeyEddi CI commands, authentication workflows, and stylistic guidelines for AI responses.
[COMMAND_EXECUTION]: While a shared utility script scripts/_skill_cli.py is included in the package, both the SKILL.md and manifest.json explicitly state that the skill uses no tools (tools: []). The script itself uses subprocess.run with a command list, which is the standard secure implementation for executing external commands when necessary, but it is not invoked by this specific skill.
[PROMPT_INJECTION]: The instructions provide operational constraints for the agent (e.g., 'Never ask if I should fix failures,' 'Only merge with authorize merge'). These are legitimate workflow definitions for the CI tool and do not attempt to bypass core AI safety filters or ignore system instructions.
[INDIRECT_PROMPT_INJECTION]: The skill defines how agents should process external data like PR review feedback. It establishes clear boundaries, such as requiring a specific 'authorize merge' phrase before performing sensitive actions, which acts as a security control against accidental or malicious merge attempts triggered by external data.

heyeddi-ci-guide