The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external data from a specific file, creating an attack surface for indirect instructions.
Ingestion points: Technical requirements are read from REQUIREMENTS.md rather than chat context.
Boundary markers: The skill mandates that requirements must be 'approved AND baselined' with Logic Gatekeeper and Human Gate 1 records before synthesis can begin.
Capability inventory: The agent can synthesize code/firmware/HDL artifacts, write build manifests, and update AI run manifests in the .agile-v/aibom/ directory.
Sanitization: The instructions explicitly require ISO 27001-compliant input validation and sanitization of all external inputs.
[DATA_EXPOSURE]: The skill incorporates strong defensive measures against credential exposure.
Evidence: The 'Secure Coding' section explicitly forbids hardcoded secrets, directing the use of environment variables or dedicated secret management systems.
[DYNAMIC_EXECUTION]: The skill allows for automated resolution of build issues within a constrained environment.
Evidence: A 'Post-Verification Feedback Loop' permits the agent to attempt to fix compilation errors or broken imports for up to three cycles, which is standard behavior for development-focused agents.

build-agent