Council
The Council extension gives you a second opinion on the active quest’s conversation. It briefs a separate agent session on what the quest has done and asks it to review the work from a red-team perspective — challenging assumptions, identifying risks and flagging blind spots.
How It Works
Section titled “How It Works”- Open the Council panel from the bottom toolbar (⌘7 by default; only one extension panel is open at a time).
- Select an agent from the dropdown: Claude Code, Gemini CLI, Codex CLI or Copilot CLI.
- Click Get Council Feedback.
Sidequest then:
- Extracts the active quest’s conversation, keeping the most recent 200,000 characters. It needs at least two messages; otherwise it asks you to let the quest run further.
- Summarizes it into a briefing with Claude. The model comes from the
councilentry in the router rules — Opus by default, since a good briefing matters more than speed here. - Saves the briefing to
~/.sidequest/council/<quest-id>.md. - Launches the selected agent in a tmux session named
sq-council-<quest-id>, in the quest’s working directory, with the briefing and a red-team review prompt. Claude Code councils run at Extra High effort and inherit the quest’s permission mode.
The agent’s live output streams into an embedded terminal in the Council panel. Reopening the panel reattaches to an existing council session; completing the quest interrupts it.
The briefing
Section titled “The briefing”The briefing is written as a self-contained dossier so the reviewer does not need the original transcript:
- Goal
- Scope & Artifacts — file paths, commits, PR and issue numbers, symbol names
- What Was Done
- Decisions & Rationale
- Assumptions
- Open Questions & Known Risks
- Verification State — what was tested and how
- Reviewer Hints
The reviewer is told to use Scope & Artifacts to inspect the actual code in the working directory rather than critiquing from the summary alone.
Red-Team Stance
Section titled “Red-Team Stance”The council agent is instructed to act as a red-team code reviewer:
- Challenge assumptions and question design decisions
- Identify potential risks, edge cases, and blind spots
- Push back on weak reasoning or unjustified choices
- Acknowledge strong decisions and good trade-offs when present
This adversarial framing helps surface issues that the primary agent may have overlooked.
Agent Selection
Section titled “Agent Selection”| Agent | How it runs |
|---|---|
| Claude Code | A separate Claude Code session in the quest’s directory |
| Gemini CLI | Google’s Gemini CLI with auto-approval |
| Codex CLI | OpenAI’s Codex CLI in full-auto mode |
| Copilot CLI | GitHub Copilot CLI with all actions allowed |
Choose the default in Settings > Model Providers > Council Agent or from the dropdown in the panel. Sidequest checks whether the selected CLI is installed and shows a detection status. OpenCode is not offered for councils.
When to Use Council
Section titled “When to Use Council”Council is most useful when:
- You want a sanity check before merging a complex change.
- A quest has been running for a while and you want fresh eyes on its approach.
- You want to compare perspectives between different AI models on the same problem.
The council session is independent of the quest session. It can read the quest’s working directory but does not interfere with the quest’s own conversation.