Deliberationv4.6.1

Commands and experts

Commands

Bundled with the Claude Code plugin (available once installed; OpenCode gets the ask-*, consensus, and analyze commands, and every other host calls the matching MCP tools directly):

Command Purpose
/deliberation:setup Configure Codex/Gemini/Grok/OpenRouter MCP servers + orchestration rules
/deliberation:help How to use Deliberation on your host, with paste-ready example prompts
/deliberation:doctor Health check (config, provider CLIs, sessions/debug, path drift) with fixes; read-only. Other hosts call the doctor MCP tool
/deliberation:dashboard Start the local read-only dashboard (runs as live state graphs, history, config, health, stats, analyzer) and print its URL; restart stops it and prints a fresh URL. Needs dashboard.enabled and a browser on the same machine
/deliberation:codex-login Log GPT (Codex) in on this machine with a ChatGPT device login - shows a link + code; for web and headless sessions
/deliberation:consensus 🔥🔥🔥 Arbiter-mediated GPT + Gemini + Grok + Claude convergence loop
/deliberation:ask-all 🔥 GPT + Gemini + Grok (+ configured OpenRouter models) in parallel, synthesized
/deliberation:ask-gpt One-shot GPT (Codex) second opinion
/deliberation:ask-gemini One-shot Gemini second opinion
/deliberation:ask-grok One-shot Grok (xAI) second opinion (advisory-only)
/deliberation:ask-openrouter One-shot OpenRouter model second opinion (advisory-only)
/deliberation:analyze Analyze recent runs (latency, tokens, verdict agreement) and suggest model/reasoning/fanout tuning (advisory)
/deliberation:uninstall Remove MCP config, rules, and aliases
/deliberation:grok-files List, prune, or gc Grok-uploaded files (storage + local cache cleanup)
/deliberation:reload-mcp Gracefully cycle the dashboard daemon and audit MCP processes after an update

/setup can also install short aliases (/ask-gpt, /ask-gemini, /ask-grok, /ask-openrouter, /ask-all, /consensus) into ~/.claude/commands/. This is opt-in. Existing same-named commands are kept by default; setup asks before overwriting any of them. /deliberation:uninstall removes an alias only if it is byte-identical to the bundled copy.

analyze is deliberately excluded: /analyze is a common name and a bare copy collides with any other plugin that ships one. Use /deliberation:analyze, which is always available. If an earlier setup installed ~/.claude/commands/analyze.md, setup now points out the path so you can delete it. /deliberation:uninstall also lists analyze, but its byte-identical guard only matches a copy of the current bundled file - an alias copied from an older release will differ and is deliberately left alone rather than deleted on a guess.

The Experts

Expert What they do Example triggers
Architect System design, tradeoffs, complex debugging "How should I structure this?" / "What are the tradeoffs?"
Plan Reviewer Validate plans before you start "Review this migration plan" / "Is this approach sound?"
Scope Analyst Catch ambiguities early "What am I missing?" / "Clarify the scope"
Code Reviewer Find bugs, improve quality "Review this PR" / "What's wrong with this?"
Security Analyst Vulnerabilities, threat modeling "Is this secure?" / "Harden this endpoint"
Researcher External libraries, docs, best practices "How do I use X?" / "Find examples of Y"
Debugger Root-cause analysis, minimal fixes "Why does this crash?" / "Debug this failing test"
Writing Editor Prose that reads like you wrote it and tells its reader what it is "Edit this launch post" / "Review the landing page headline and lede"

When experts help most

  • Architecture decisions - "Should I use Redis or in-memory caching?"
  • Stuck debugging - after two or more failed attempts, get a fresh perspective
  • Pre-implementation - validate a plan before writing code
  • Security concerns - "Is this auth flow safe?"
  • Code quality - a second opinion on your implementation

When not to use experts

  • Simple file operations (Claude handles these directly)
  • First attempt at any fix (try yourself first)
  • Trivial questions (no need to delegate)

Writing in your own voice

The Writing Editor runs whenever your agent writes, reviews, or updates text you publish under your name: posts, articles, newsletters, release notes, README intros, landing pages and launch copy, PR descriptions, bios, talk abstracts. Your agent drafts first, then the editor checks the draft before you see it. It returns line edits, questions about anything the text needs that only you know, and a rewrite that keeps every fact. On product, landing, launch, and README-intro copy it also runs a reader test: the headline and the line under it must say what the thing is and who it is for, and the first ~50 words what changes for the reader and what to do next. Strings you pinned stay verbatim in the rewrite; when one fails the test, the editor still reports it under "Author decisions" with a proposed replacement. Name the audience when you ask; without one, the editor states the reader it assumed. For a piece that matters, run /consensus on the draft and the panel reviews it with the writing-editor persona.

The editor needs evidence of how you write. Without it, it can only cut filler. A short voice file helps most. Keep it in your repo and point to it ("use my voice file at docs/voice.md"):

markdown
# <Your name> - voice

## How to use this file
- These rules beat any default style rule.
- Samples show rhythm only. Do not copy their phrases.
- Never invent a fact, number, person, customer, or quote.

## What I argue from
- 3-7 beliefs you keep coming back to, in your words.

## Mechanics
- "I" or "we", contractions or full forms, sentence length mix.
- Name the specific thing. A number beats an adjective, and every number needs a source.
- Typography you use (dashes, quotes, heading case).

## Moves
| Move | Use only when | My own example |
|------|---------------|----------------|
| <a habit of yours> | <the condition> | "<a real line you wrote>" |

At most one move per short post.

## Would sound wrong
| Wrong | Why |
|-------|-----|
| "Thrilled to share" | Launch voice. Say what shipped. |
| "Here's the thing", "Sound familiar?" | Run-up. Start with the thing itself. |
| A one-line quotable closer | End on the last concrete fact. |

## Channels
- LinkedIn: the hook fits before the "see more" cut, 1-3 sentence paragraphs, no closing question.
- Blog: ...

## Samples (rhythm only)
- 5-10 short lines you wrote yourself: posts, replies, talk transcripts.

Two parts of the file do most of the work. Each move has a condition, so the editor does not paste your signature line into every post (overused habits read as machine-made too). The "would sound wrong" table lists phrases you would never write. When you edit a draft by hand, add the voice change you made to the file as a rule.

Picking models:

  • Review with several model families. /consensus with the writing-editor persona sends the draft to every voice in your consensus panel. Reviewers from different families tend to flag different tells (one catches the arc of the piece, another a claim you never made).

  • Give rewrites their own model record. A record restricted to writing-editor joins only prose reviews, so its settings do not touch your code reviews. Use a higher temperature for writing than the 0.2 typical for review. Pick the model by reading what it writes on your own samples, not by its benchmark scores:

    json
    "models": {
      "prose": {
        "provider": "openrouter",
        "model": "<provider/model whose prose you like>",
        "experts": ["writing-editor"],
        "consensus": true,
        "temperature": 0.7
      }
    }
  • Gemini is the only provider that can write the rewrite back to the file (implementation mode). The others return it as text.

The editor does not score text for "AI likelihood" and does not promise any detector result. It also never adds typos or broken grammar to make text look human.

How to Use

Describe your task. Your agent detects when an expert helps and delegates automatically (the examples show Claude Code):

text
You: "Is this authentication flow secure?"
Claude: routes to the Security Analyst, then synthesizes the findings.

You can also ask explicitly: "Ask GPT to review this architecture", "Ask Gemini to...", or "Ask Grok to...". Each expert runs read-only for analysis; Gemini can also run with write access to apply fixes, and your agent picks the mode from your request.

Or invoke the slash commands directly - see Commands above.

Rendered from README.md in the repository.

Type to search the site.