The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local JavaScript files (audit-esteira.mjs and validate-mermaid.mjs) using the node runtime. This is a functional requirement for the skill's auditing purpose.
[PROMPT_INJECTION]: The skill processes project data such as specifications, task lists, and state files, which creates a surface for indirect prompt injection. Malicious instructions within these files could potentially influence the agent's evaluation results.
Ingestion points: Project files including tasks.md, STATE.md, context-map.md, and various markdown files in the specs/ directory.
Boundary markers: Absent; the instructions do not specify delimiters or warnings to ignore instructions embedded in the audited data.
Capability inventory: Shell command execution via node for structural validation.
Sanitization: Absent; the skill relies on the agent's native reasoning to identify violations without explicit input validation.

auditar