The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The 'review' mode of the skill fetches and processes GitHub Action run logs using the gh run view --log command. This represents a surface for indirect prompt injection where malicious instructions could be embedded in build logs to influence the agent's audit report or drive unauthorized API actions.
Ingestion points: The skill executes gh run view "$id" --log to inspect logs for cache hit rates and error diagnostics in SKILL.md.
Boundary markers: No specific delimiters are used to separate log data from agent instructions.
Capability inventory: The agent uses the gh CLI for API calls, reactions, and comments, and for listing workflow runs.
Sanitization: The skill does not implement filtering for retrieved log content.
[DYNAMIC_EXECUTION]: The skill generates CI workflow templates that utilize inline language interpreters to process data files during the build process.
Evidence: Templates in templates/node-ci.yml.md and templates/reusable-workflow.yml.md execute node -e to parse JSON coverage data. The templates/python-ci.yml.md template executes python -c to process XML test results.

github-actions-author