The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a multi-agent evaluation pipeline that processes model-generated text.
Ingestion points: The skill processes output from multiple seats in answers.json and judge rankings in judges.json.
Boundary markers: Implements a blind-ranking mechanism to prevent identity-based bias and utilizes html.escape during report generation.
Capability inventory: Executes a local script (harness/scripts/council.py) and orchestrates tasks via the orca orchestration binary.
Sanitization: The skill instructions explicitly mandate the use of html.escape to mitigate HTML injection risks in the final reports.
[COMMAND_EXECUTION]: The skill coordinates model evaluations by executing a local Python script and utilizing the orca orchestration tool to dispatch model calls.
[EXTERNAL_DOWNLOADS]: References external repositories github.com/karpathy/llm-council and github.com/0xNyk/council-of-high-intelligence as authoritative sources for logic and persona data. These references are documented neutrally as technical context.

council