The Agent Skills Directory

[PROMPT_INJECTION]: The skill reads potentially untrusted instructions from external SPEC.md files and incorporates them into the task definitions for implementation agents, creating a risk of indirect prompt injection.
Ingestion points: Phase 1 (Intake) reads the provided SPEC.md file content.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are applied when extracting text from the specification.
Capability inventory: The skill can write to the local filesystem and execute internal scripts.
Sanitization: Input content is triaged for structure but is not sanitized for potential malicious instructions before being passed to downstream agents.
[COMMAND_EXECUTION]: The skill executes a local schema validation script using the bun runtime. This execution is confined to the skill's own directory and a related local 'implement' skill path.

decompose