The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains explicit instructions to operate in "AUTONOMOUS MODE" and commands the agent to "Do NOT ask questions" and "Do NOT pause for confirmation." These instructions are designed to bypass standard agent safety guardrails and eliminate human-in-the-loop oversight.
[COMMAND_EXECUTION]: The skill executes various local CLI validation commands (e.g., terraform validate, docker build --check, kubectl dry-run) during its self-healing phase. These commands are triggered based on the content of the project being analyzed.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to its processing of untrusted project data.
Ingestion points: Manifest files including package.json, requirements.txt, go.mod, and Cargo.toml from the current project directory.
Boundary markers: None identified. The skill directly interpolates detected scripts and package names into generated YAML workflow files.
Capability inventory: The skill has the ability to write files to the .github/workflows/ directory and execute shell commands for validation.
Sanitization: No sanitization or validation of the ingested strings (like script names or framework identifiers) is performed before they are placed into executable CI/CD configurations.
[DATA_EXPOSURE]: The skill attempts to write execution metadata and telemetry to a specific path on the user's filesystem (~/.claude/projects/).

github-actions