The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/test_assets.py utilizes subprocess.run to invoke the bundled Python handler assets/shared/observe.py for automated validation of hook fixtures. This execution is confined to the local environment and the current Python interpreter, serving only to verify that configuration assets are correctly parsed.
[INDIRECT_PROMPT_INJECTION]: The skill defines a standard mechanism for agents to process lifecycle events, creating a potential surface for indirect injection from provider-generated data. 1. Ingestion points: assets/shared/observe.py ingests data from standard input (sys.stdin). 2. Boundary markers: The handler script strictly validates that incoming data is a valid JSON object. 3. Capability inventory: Command execution is used in the local test suite; the configuration assets provided as templates direct various agent harnesses to invoke the handler script. 4. Sanitization: Data is parsed using the standard json library, and the handler returns an empty object without performing any further actions on the input, effectively neutralizing potential injection attempts.
[SAFE]: The skill instructions provide security-focused guidance, explicitly warning users that agent hooks represent trust boundaries and should be reviewed for inherited secrets and filesystem effects. All external links and package references target official documentation and libraries from established technology providers.

maintain-agent-hooks