The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted configuration data from the .agents/runtime-profile.yml file located in project repositories. A maliciously crafted profile could attempt to influence the agent's behavior or trigger unintended command execution.
Ingestion points: .agents/runtime-profile.yml (referenced in SKILL.md and references/runtime-profile.md).
Boundary markers: The instructions advise the agent to "Read it before acting; do not invent missing commands" and explicitly state that "Command presence is capability, not authorization."
Capability inventory: The skill facilitates the execution of shell commands defined in the profile's runtime.commands section, such as status checks and environment setup.
Sanitization: The underlying CLI tool tokenizes commands to prevent shell injection and performs schema validation on the profile.
[COMMAND_EXECUTION]: The skill directs the agent to execute shell commands for environment management. This includes vendor-provided CLI commands (e.g., de-novo skills validate) and project-specific commands (e.g., docker compose up, node tools/dev-overlay.mjs) defined in the repository's profile.
[EXTERNAL_DOWNLOADS]: The documentation instructs the user to install the associated CLI tool using npm install && npm link within the vendor's catalog checkout.

grove