The Agent Skills Directory

[COMMAND_EXECUTION]: Executes the gh pr list command to retrieve open pull requests from the current repository. This is a standard use of the GitHub CLI to anchor the agent's work to existing code reviews and development tasks.
[DATA_EXPOSURE]: Accesses project-specific local files including memory/MEMORY.md, soul/SOUL.md, and memory/cron-state.json. These files contain project context, goals, and identity, which are used to generate relevant actions. No access to system-level sensitive files (e.g., SSH keys, AWS credentials) was detected.
[PROMPT_INJECTION]: The skill processes untrusted input from external sources such as pull request titles and topic files to extract "loops." While this represents an indirect prompt injection surface, the risk is significantly mitigated by the skill's multi-step validation process, which includes a "specificity gate" and a "banned-phrase lint" to ensure the resulting actions are concrete and objective.

action-converter