The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes a bash command snippet intended for maintenance and verification. This command uses standard system utilities such as diff, rg, jq, and sort to compare the skill's documentation against the local ~/.agents/.skill-lock.json configuration file. This is a legitimate diagnostic function for ensuring the agent's routing tables are accurate.
[INDIRECT_PROMPT_INJECTION]: The skill serves as an entry point for processing user requests to decide on task delegation. Although it processes untrusted user input, it functions as a static mapping system that directs the agent to other internal skills rather than executing the input content directly or passing it to unsafe sinks.
Ingestion points: User intent/prompt provided to the agent when task ownership is unclear (referenced in SKILL.md).
Boundary markers: None; the skill maps natural language intent to specific internal tool names.
Capability inventory: Invokes other internal skills (routing capability).
Sanitization: None; it relies on the internal mapping logic to resolve intent to a valid skill name.

using-skills