From consensus-loop
Orchestrates consensus-loop sessions: reads handoffs, identifies unblocked tasks, distributes to parallel workers/agents, tracks assignments, manages corrections via SendMessage.
How this skill is triggered — by the user, by Claude, or both
Slash command
/consensus-loop:orchestratorThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
You are the orchestrator. You do NOT implement — you distribute, verify, and decide.
You are the orchestrator. You do NOT implement — you distribute, verify, and decide.
Read the corresponding reference when entering each phase:
| Phase | Reference | When |
|---|---|---|
| Task complexity tiers | references/tiers.md | Before spawning — evaluate Tier 1/2/3 |
| Scout / RTM generation | references/scout.md | Before distributing Tier 2/3 work |
| Multi-track distribution | references/distribution.md | When spawning parallel workers + track closure |
| Correction cycle | references/correction.md | On [pending_tag] rejection + upstream delays |
| Retro / merge / lifecycle | references/lifecycle.md | After [agree_tag] + session end audit |
| Context | Detection | Behavior |
|---|---|---|
| Interactive | Main session, user present | Present options → wait for selection → execute |
| Headless | Subagent, no human | Auto-select unblocked tasks → execute → report |
In headless mode, NEVER ask questions. Auto-select based on dependency order, auto-block on escalation triggers, write session summary to file.
Read config: ${CLAUDE_PLUGIN_ROOT}/config.json
consensus.watch_file → evidence file pathconsensus.planning_dirs → design document directoriesplugin.respond_file → auditor verdict fileplugin.handoff_file → session handoff path (default: .claude/session-handoff.md)session-start.mjsagent_id) → present resumption optionsTrack agent assignments in the handoff file:
### [task-id] Task Title
- **status**: not-started | in-progress | auditing | correcting | done
- **depends_on**: other-task-id | —
- **blocks**: other-task-id | —
- **agent_id**: <agent-id>
- **worktree_path**: <path>
- **worktree_branch**: <branch>
Registry rules:
agentId, worktreePath, worktreeBranch in handoffagent_id — never spawn new agentdone, keep agent fieldsSendMessage to resume in-progress tasksSession Start
↓
Evaluate Tier → read references/tiers.md
↓
┌─ Tier 1 (Micro): direct fix → verify CQ+T → commit → next task
├─ Tier 2 (Standard): scout? → worktree → audit cycle → retro → merge
└─ Tier 3 (Complex): mandatory scout → worktree → full audit → post-merge regression → retro
↓
Result Verification
├─ [agree_tag] → Retro & Merge → read references/lifecycle.md
└─ [pending_tag] → Correction → read references/correction.md → loop
↓
Write Handoff → next task → loop
${CLAUDE_PLUGIN_ROOT}/templates/references/${locale}/done-criteria.md${CLAUDE_PLUGIN_ROOT}/templates/references/${locale}/evidence-format.mdsubagent_type: "consensus-loop:implementer", isolation: "worktree", run_in_background: truenot-started → in-progressWhen worker completes:
[agree_tag] → proceed to Retro & Merge (read references/lifecycle.md)[pending_tag] → Correction Cycle (read references/correction.md)When a task requires new track definition:
/consensus-loop:planner with the requirementBefore spawning, verify:
depends_on completedBlocked → skip → select next unblocked.
agent_idreferences/distribution.md)references/lifecycle.md)npx claudepluginhub berrzebb/claude-plugins --plugin consensus-loopOrchestrates multi-agent parallel execution for complex tasks like features, refactoring, testing, reviews, and documentation using cc-mirror tracking and TodoWrite visibility.
Orchestrates parallel multi-agent Evaluate-Loop v3 workflows for code tracks. Handles /go goals, worker dispatch, board deliberation, message bus monitoring, and metadata.json state tracking.
Creates structured wave execution plans with role-based agent assignment, dependency ordering, and checkpoints after user alignment. Activated automatically after session-start Q&A phase.