The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied markdown files (Brainstorm, Define, Design) which contain implementation requirements and verification commands, creating an attack surface for indirect prompt injection.
Ingestion points: The skill reads BRAINSTORM_{FEATURE}.md, DEFINE_{FEATURE}.md, and DESIGN_{FEATURE}.md files to extract requirements, verify gates, and file manifests.
Boundary markers: The instructions include a "Drift detection" step to check for unauthorized changes or reintroduction of rejected items, though this is an instructional rather than a technical boundary.
Capability inventory: The skill is designed to modify project source code and execute shell commands specified in the project's metadata (e.g., the verify_gate command).
Sanitization: There is no evidence of technical sanitization for the interpolated specification content; the skill relies on the agent's adherence to the defined protocol and user confirmation before the first alteration.
[COMMAND_EXECUTION]: The skill executes various local shell commands to validate the implementation and the final report.
Executes a bundled script, scripts/validate_build_report.py, to verify the structure and consistency of implementation reports.
Invokes project-specific scripts like scripts/verify-gate.sh or arbitrary shell commands defined within the DEFINE specification file.
Performs standard development verification tasks such as lint, typecheck, and automated tests as specified in the project configuration or design.

sdd-build