The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates the execution of various developer tools for project validation.
Evidence: Recommendations to run tsc, eslint, gitleaks, and semgrep.
Evidence: Provision of a reference script floor-guard.mjs in references/floor-guard.md that uses child_process.execFileSync to run git commands.
[EXTERNAL_DOWNLOADS]: The skill recommends installing several well-known developer tools from official registries.
Evidence: Instructions to install semgrep, gitleaks, osv-scanner, lighthouse, and axe-core via NPM, PyPI, or Homebrew.
[INDIRECT_PROMPT_INJECTION]: The skill processes external file data to guide agent behavior.
Ingestion points: CONSTRAINTS.md and git diff output.
Boundary markers: Not explicitly defined for the file content, but the skill includes instructions to watch for bar-loosening regressions.
Capability inventory: Shell command execution via execFileSync and package manager scripts.
Sanitization: Emphasizes the use of gitleaks --redact to prevent secret exposure in transcripts.
[DATA_EXFILTRATION]: The skill includes explicit security measures to prevent credential leaks during analysis.
Evidence: Warning in SKILL.md that --redact on gitleaks is mandatory to avoid secrets landing in the agent's transcript or logs.

constraint-driven-development