The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted specification files from the .specs/ directory to guide its implementation and testing logic.
Ingestion points: Reads .specs/SPEC-{YYYYMMDD}-{feature}.md files to identify requirements and task plans.
Boundary markers: None identified. The instructions do not specify delimiters or warnings to ignore instructions embedded within the SPEC content.
Capability inventory: The skill is instructed to write implementation code and tests, and then execute them using shell commands for test suites, linting, and type checking.
Sanitization: The scripts/validate_spec.py script validates the status and structure of the file but does not sanitize the natural language content for malicious instructions.
[DYNAMIC_EXECUTION]: The skill generates and executes code and test scripts at runtime based on the interpreted specifications.
Evidence: The Red-Green-Refactor loop in SKILL.md involves writing code to files and immediately executing test suites.
[COMMAND_EXECUTION]: The agent is prompted to run various system commands (tests, lint, type checks) which are determined by the project stack and the specifications being processed.

execute-specs