The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on onboard.py for project scanning and registration. The instructions explicitly direct the agent to run python3 onboard.py --refresh monthly and python3 onboard.py --scan. While the script performs read-only operations during scanning, the agent is encouraged to execute shell commands based on its output.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from multiple sources in a project directory (e.g., package.json, Makefile, CONTRIBUTING.md, and test file contents) to infer 'Key Invariants' and 'Conventions'.
Ingestion points: Reads project files like package.json, Makefile, CONTRIBUTING.md, and directory structures in onboard.py and SKILL.md Step 1 & 2.
Boundary markers: None. The skill does not implement delimiters or warnings to ignore instructions embedded within the files being scanned.
Capability inventory: The skill has the ability to write files (PROJECT.md) and execute Python scripts (onboard.py), which in turn can influence the agent's long-term behavior for that project.
Sanitization: None. The detected strings from package.json scripts or Makefile targets are directly interpolated into the PROJECT.md context file.
[SAFE]: The use of standard configuration files (YAML, JSON) for state management and the reliance on standard project indicators are best practices for this use case and do not escalate the verdict.

project-onboarding