The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the installation of external dependencies, specifically the @intentsolutions/jrig-cli package, which is a vendor-owned resource managed by the skill's author. It provides instructions for global and per-repository installation via npm and pnpm.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute git operations and the custom j-rig binary. This is used for skill evaluation, proposal application, and status tracking.
[PERSISTENCE]: The skill implements a multi-layered hook architecture (Sinker, Line, Hook) that monitors tool-use events within the agent's environment. The L3 layer persists by intercepting git commit and git push commands to enforce quality gating, which can include blocking the execution of these commands via non-zero exit codes.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill reads and processes the content of SKILL.md files located in user-provided skill directories.
Boundary markers: The instructions do not define specific delimiters or "ignore" commands to protect the refiner's internal prompts from instructions embedded within the skill files being analyzed.
Capability inventory: The skill possesses Read, Write, Edit, Glob, and Bash capabilities, which are used to modify files and execute project-level commands.
Sanitization: While the process includes human-in-the-loop validation for final promotion and a binary evaluation score for acceptance, the initial ingestion of untrusted markdown content for scoring and proposal generation lacks explicit sanitization or escaping mechanisms.

j-rig