council-cli
A visible review ritual for multi-agent technical critique.
A command-line roundtable that turns model disagreement into structured planning, architecture review, and implementation critique.
A shared prompt context is passed to specialized model roles.
Each role contributes critique from a distinct angle such as architecture, risk, implementation, or verification.
A synthesis step produces the decision record and next action.
Context
Coding agents are useful, but their reasoning can be brittle when a task needs architectural pressure testing.
council-cli explores a lightweight ritual where multiple perspectives critique the same problem before the final direction is chosen.
Constraints
The workflow has to avoid endless debate.
Roles must be distinct enough to create real disagreement.
The final artifact has to be actionable, not just a transcript.
Tradeoffs
Kept the interface CLI-first to reduce UI overhead while testing the workflow.
Favored repeatable review structure over fully autonomous implementation.
Accepted extra ceremony for decisions where architectural risk is real.
Outcomes
Produced a reusable model for structured AI-assisted review.
Strengthened the broader agentic-systems portfolio by showing how orchestration can support judgment, not just speed.
Lessons
Good review workflows make disagreement legible.
Decision support is strongest when the output records why alternatives were rejected.