ensembra

Ensembra

Where agents perform in concert — a multi-agent orchestrator plugin for Claude Code.

Screenshots

/ensembra:run feature — 5-phase pipeline with consensus and reuse evaluation	/ensembra:config — interactive picker with cascade-safe custom mode
/ensembra:transfer — 10-section handover document with devils-advocate pitfalls

What is Ensembra?

Ensembra is a Claude Code plugin that orchestrates six specialist agents and one scribe through a 5-phase pipeline to produce structured code reviews, mutual supervision, automatic documentation, and project handover documents. Built for solo developers who want team-level deliberation without the team.

Key ideas:

• Separation of deliberation and execution: external LLMs (Ollama / Gemini) debate, Claude Code executes.
• Reuse-First cross-cutting policy: four toggleable devices force every performer to consider existing code before writing new code.
• Deep Source Inspection: 10-item checklist (6 forced + 4 optional) prevents shallow reads.
• Consensus-driven flow: 70/40 thresholds gate Phase 2 execution and halt pipeline on strong disagreement.
• Automatic documentation: every task produces a report; weekly roll-ups and handover documents are built-in.

5-Phase Pipeline

Phase 0 Gather     — Deep Scan produces a shared context snapshot
Phase 1 Deliberate — R1 → (optional R2) → Synthesis with peer signatures
Phase 2 Execute    — Claude Code edits files per the agreed plan
Phase 3 Audit      — designated performers verify the diff
Phase 4 Document   — scribe records Task / Design / Request / Daily / Weekly

Performers (v0.8.0+ Debate/Audit split)

v0.8.0 splits Performers into three tiers. Opus is forbidden in debate and used only by the final-auditor in Phase 3 — the unanimous-consensus judge.

Debate (Phase 1) — external LLMs + sonnet-or-lower, no opus:

Role	Responsibility	Default Transport	Default Model
🧭 planner	Requirements, acceptance criteria	Claude sub-agent	sonnet (v0.8.0: was opus)
🏛 architect	Module boundaries, patterns	MCP (Gemini) → Ollama → Claude	gemini-2.5-flash
🛠 developer	Implementation strategy	Claude sub-agent (opt-in external chain)	sonnet
🛡 security	Threats, secrets, OWASP	Ollama → Claude	qwen2.5:14b
🧪 qa	Edge cases, regression	Ollama → Claude	qwen2.5:14b
😈 devils-advocate	Counter-arguments, YAGNI	Claude sub-agent	haiku

Audit (Phase 3) — specialist auditors → final-auditor:

Role	Responsibility	Default Transport	Default Model
(specialists)	Preset-specific specialists (see presets table)	varies	varies
⚖️ final-auditor	Unanimous-consensus judge, always last (v0.8.0)	Claude sub-agent	opus

Document (Phase 4):

Role	Responsibility	Default Transport	Default Model
✍️ scribe	Phase 4 documentation	Claude sub-agent	sonnet

Unanimous consensus (v0.8.0) = Phase 1 agreement ≥ 70% AND final-auditor.verdict == pass. Final-auditor rework is capped at 1 cycle (opus cost control). See CONTRACT.md §8.8 for the generalized Transport Fallback Chain Protocol and §11.3 for Final Audit details.

All models auto-fall back to Claude sub-agents when the external transport is unavailable.

Skills

• /ensembra:run <preset> <request> — main pipeline entry point
• /ensembra:config — unified interactive settings picker (all options, all cascade-safe)
• /ensembra:transfer [scope] — project handover document (full project, path, or natural-language scope)
• /ensembra:report daily|weekly — roll-up reports

Presets