agentleFS
Sign inStart free
spscreamspscream/ai-skills/skills/consensus/SKILL.md

consensus

Skill1 starsChanged 3 months agoMITView on GitHub

Use when a decision is expensive to reverse and one model's answer is not enough — architecture splits, storage or codec choices, a design the user doubts, a claim that sounds too confident. Also use when the user asks for a second opinion, another model, other models, GPT, Grok, Gemini, консенсус, второе мнение, спроси другие модели, что скажут другие.

  • Installs packages

3 other projects use a version of this.

Adopted by

Most recently taken up by

  • Other versions3 projects use a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 3

    1. ibahgat/oh-my-iflow★ 2same name
    2. github-wyy/consensus★ 1same name
    3. spsk-dev/consensus★ 0same name

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

  • In this repository2 other files it gives its agents.

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 2 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Run scripts/consensus.sh from this skill directory.

    line 37
  2. Distinctive:

    Do not write "our chat".

    line 72
  3. Distinctive:

    Write the synthesis in the language that the user uses.

    line 90

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Consensus of several models

A panel of models from different vendors answers one question. Then each model reads the other answers and revises its own. The panel runs on the Claude Code and Cursor subscriptions. It needs no API key.

The disagreements carry the value. The agreement does not. The models learn from overlapping data, so they make the same mistake together. Never present agreement as proof.

When to use the panel

  • The decision is expensive to reverse: an architecture split, a storage choice, a training setup.
  • You have a recommendation, but you are not sure. A wrong choice costs days of work.
  • The user asks for a second opinion. The user asks what other models say.

When not to use the panel

  • You can measure the fact. You can run a test. You can read the documentation. Measure instead.
  • The task is mechanical: a change by example, a rename, a known bug.
  • You need speed. One run takes about three minutes and six full answers from the quota.

How to run the panel

The question is about code in a repository. Then always use -d. This is measured: the panel with -d found the precedent that the project set already. The panel without -d answered from the textbook and missed it.

Run scripts/consensus.sh from this skill directory. Claude Code installs the skill under ~/.claude/skills/consensus, so the full path is ~/.claude/skills/consensus/scripts/consensus.sh.

# Repository, read-only. This is the default form:
scripts/consensus.sh -d "$PWD" -q "вопрос" -r 2

# Without -d the panel sees only the files you pass. Use this form for sensitive code:
scripts/consensus.sh -q "вопрос" -f path/to/file -r 2

Set the Bash timeout to 900000 ms or more. Run -h for the flags and the default panel.

Requirements

  • claude — the Claude Code CLI, for a claude: participant.
  • agent — the Cursor CLI, for a cursor: participant. Log in with agent login.
  • jq — the script reads the Cursor answer from JSON with it.
  • timeout or gtimeout — optional. Without either one the model calls run with no time limit and the script says so on stderr. macOS ships neither until brew install coreutils.

A participant whose CLI is missing gets no answer, and the script says so for that participant only. A panel of one harness is legitimate: -m claude:fable -r 1 is a single reader with read-only access to the repository, which is an audit rather than a consensus.

Both CLIs run on their own subscription. The panel needs no API key. The Cursor model ids depend on the account. List your own with agent models. Set CONSENSUS_PARTICIPANTS to change the default panel.

How to write the question

The models know nothing about this conversation. They know nothing about the project. The question must read on its own.

  1. Name the stack, the volumes, and the limits. Write "Elixir/OTP, up to 50 thousand members". Do not write "our chat".
  2. Name the entry points: the modules, the directories, the tests. A model that gets no entry point reads the wrong files in a large tree.
  3. Tell the panel what you tried already. Tell the panel why it did not work.
  4. Limit the length of the answer. Ask each model to name the main risk of its own choice. The question about risk shows the difference between two models that chose the same option.

How to write the synthesis

You write the synthesis. The script prints the material and stops.

  • Agreement — one line for each point. Do not retell the answers.
  • Disagreement — the main part. Name who claims what. Name the basis of each claim: a measurement, the documentation, or an assumption.
  • What would settle it — the measurement or the source that closes the question. Give the exact command, if you know it.
  • Your own recommendation — with a reason. Do not report the majority as the reason.

Write the synthesis in the language that the user uses.

A participant gave no answer. Then say so. Do not give that participant an opinion.

Common mistakes

MistakeDo this instead
A panel of models from one vendorUse different vendors. One vendor gives false agreement
A question with "our" and "this"Write the full question. It must read without the session
A synthesis that retells three answersAgreement, disagreement, what settles it, recommendation
"All three agree, so it is correct"The training data overlaps. Agreement is not proof
A panel instead of a measurementMeasure first. Use the panel for a judgement

Safety

Both harnesses run read-only: Claude without Edit, Write and Bash, Cursor with --mode ask. This is tested — each one reads a file, and each one refuses to create a file. The Cursor flag --sandbox enabled does not block a write. Do not treat it as protection.

With -d the panel sees the whole repository, gitignored files included. For sensitive code, drop -d and pass the files with -f.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

Check your setup, free