The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes pnpm scripts and git commands for repository discovery and test execution. These are standard operations for the skill's stated purpose of managing Paperclip evaluations.
[INDIRECT_PROMPT_INJECTION]: The skill interprets and reports on data from repository files and evaluation evidence, which could contain instructions designed to influence the agent.
Ingestion points: Evaluation catalogs, repository documentation, and fixture conversations processed by the agent.
Boundary markers: Relies on instructional constraints directing the agent to follow established projection and publisher checks.
Capability inventory: File system read access and execution of pnpm and git commands within the repository.
Sanitization: Explicitly instructs the agent to use sanitized projections and allowlisted tool outcomes to prevent exposure of raw artifacts, secrets, or credentials.

paperclip-evals