The Agent Skills Directory

[COMMAND_EXECUTION]: The skill generates a bash script (.claude/hooks/quality-gate.sh) and configures the AI agent's environment (.claude/settings.json) to execute this script as a 'Stop' hook. This script performs local file checks and runs git diff to identify hardcoded secrets or staged .env files. This is a platform-supported feature used here for quality assurance and security gating.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by scanning local project files such as README.md or package.json to automatically infer team roles and define agent personas.
Ingestion points: Reads local project files (README.md, package.json, pyproject.toml, etc.) during Phase 0A.
Boundary markers: None explicitly defined during the scanning and inference phase.
Capability inventory: File system writes (creating .claude/ directories and files), shell script generation, and configuration of execution hooks.
Sanitization: The skill uses the inferred context to populate templates; no specific sanitization or escaping of the ingested file content is mentioned before it is written into new configuration files.

aos