The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs state transitions by executing a local script node scripts/spec-superflow.mjs. It also performs syntax checks using node --check on JavaScript/ESM files. These are standard operations for a development automation tool.
[DATA_EXPOSURE]: The agent reads local project files, including execution-contract.md, tasks.md, and architectural decisions in design.md, to verify that requirements have been met. This data access is restricted to the local file system.
[PROMPT_INJECTION]: The skill employs a 'Verification Iron Law' and 'Gate Function' to constrain the agent's behavior. These are defensive prompt engineering techniques used to ensure the agent does not provide false status reports, rather than malicious attempts to bypass safety filters.

closure-archivist