The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands via the claude CLI and an external tool called air. It also describes configuring claude hooks to execute local scripts like ./.claude/hooks/observer.mjs during the session, which creates a risk of executing code found within an untrusted repository.
[EXTERNAL_DOWNLOADS]: References an external, unverifiable third-party tool named "AirAgent" (air command) for managing task turn budgets.
[PROMPT_INJECTION]: The skill processes repository source code, which serves as a vector for indirect prompt injection.
Ingestion points: Source code and project files located at the user-specified absolute path.
Boundary markers: Recommends using explicit "Write scope" definitions and "result contracts" to limit agent actions and expectations.
Capability inventory: Employs the claude CLI with broad tool access including Read, Edit, Write, and Bash execution.
Sanitization: Lacks automated input sanitization, instead advising human inspection and local validation of all changes and reports.

claude-code-proper