The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a suite of bundled Python scripts to perform artifact validation and self-testing. These scripts are invoked via subprocess.run (e.g., in scripts/selftest.py) using direct argument lists without shell interpolation. These executions are localized to the skill's own internal directory to maintain process integrity during validation phases.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied markdown artifacts to manage feature development cycles.
Ingestion points: Markdown files such as plan.md, checks.md, and verification.md located within the project's .specs/features/ directory.
Boundary markers: The process uses rigid markdown headers and EARS (Easy Approach to Requirements Syntax) keywords to delimit instructions and obligations.
Capability inventory: The skill can perform file system writes to create/update artifacts and execute local validation scripts.
Sanitization: Deterministic Python validators (validate_plan.py, validate_checks.py) are used to enforce structural integrity, mandatory fields, and precise phrasing, significantly reducing the surface area for indirect injection through malformed artifact content.
[PERSISTENCE]: The skill's documentation and scripts (check_commit.py) describe an optional integration for a git commit-msg hook to enforce Conventional Commits. This involves creating a symbolic link in the .git/hooks directory, which is a standard development workflow improvement that requires explicit user opt-in.

tlc-spec-lean