The Agent Skills Directory

Multi-Model Consensus Mechanism: The skill implements a robust review process by spawning independent subagents using different AI models. This design is intended to reduce individual model bias and improve accuracy through consensus synthesis.
Read-Only Scope Enforcement: The instructions explicitly mandate that subagents operate in a read-only mode for the research phase, citing exact files and lines without speculating or suggesting unsolicited patches.
Standard Development Workflow: The optional 'review-and-fix' mode follows standard development practices, presenting findings to the user before applying fixes and using subagents for targeted investigations.
No Code or Executables: The skill consists entirely of markdown instructions and evaluation configuration files. It does not include scripts, binaries, or external dependencies, which significantly minimizes its attack surface.
No Network or Sensitive Access Findings: The skill does not perform network operations to non-whitelisted domains or access sensitive system paths (e.g., credentials, SSH keys). Its data ingestion is restricted to the code context provided by the user or the git environment.

council-review