The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions in SKILL.md (Step 3: Validate It) direct the agent to execute shell commands such as make setup, make dev, and make test using the Bash tool to verify templates that the agent itself generated.
[PROMPT_INJECTION]: The skill processes user-provided codebase content (Step 0: Friction Audit), creating a surface for indirect prompt injection. • Ingestion points: Local project files, existing service structures, and CI configurations accessed via Read, Grep, and Glob tools. • Boundary markers: The instructions do not specify delimiting or ignoring embedded instructions in the audited code. • Capability inventory: The skill has access to Write, Edit, Bash, and WebFetch tools, which could be abused if the agent follows malicious instructions found in the user's codebase. • Sanitization: No explicit sanitization or validation of the ingested code structure is mentioned.
[EXTERNAL_DOWNLOADS]: The skill instructions include examples and validation steps that involve downloading and running external packages or templates (e.g., npx create-myapp, cookiecutter) which may fetch content from remote repositories during the template execution phase.

pave-golden