The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to evaluate plans and work outputs generated by other agents. This creates a vulnerability surface where malicious instructions could be embedded in the data being reviewed to influence the agent's verdict or override the review checklist.
Ingestion points: Agent plans, work outputs, and completion claims reviewed in SKILL.md.
Boundary markers: None specified; the skill processes the full content of the items being reviewed.
Capability inventory: The skill itself defines a checklist but relies on the agent's general capabilities (file reading, command execution) to perform the review.
Sanitization: No explicit sanitization or instruction-filtering for the reviewed content is defined.
[COMMAND_EXECUTION]: The global rules in AGENTS.md instruct the agent to use the compose-agentsmd tool for managing rule sets. This involves executing commands to synchronize rules and push changes to GitHub repositories. These operations are part of the vendor's standard workflow for maintaining agent instructions.
[EXTERNAL_DOWNLOADS]: The documentation and rules suggest installing the compose-agentsmd tool via npm or running it via npx. These resources are associated with the skill's authoring organization and are used for instruction management.

user-proxy