The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates the execution of local shell commands, tests, and benchmarks (referred to as 'oracles') to settle architectural decisions and verify candidate claims. This is a primary functional requirement of the skill and is constrained by instructions to verify evidence independently.
[PROMPT_INJECTION]: The skill ingests external code and sub-agent outputs, creating an attack surface for indirect prompt injection. Ingestion points: Reads repository files and sub-agent JSON responses. Boundary markers: Uses structured JSON blocks for all inter-agent communication. Capability inventory: File reading, sub-agent spawning, and shell command execution ('re-run it'). Sanitization: Implements mandatory schema validation and a 'Hardened judge contract' that requires independent re-reading of source files to verify any claim made by potentially compromised or hallucinating candidates.

work-judge