duo
Skill22 starsChanged 41 days agoMITView on GitHub
Run a traceable multi-model session between Claude (the user's Claude plan) and OpenAI Codex (the user's ChatGPT plan) with the duo harness - debates that converge on a checked claim ledger, cross-validated code reviews, anonymized councils, or parallel asks, with every model and effort chosen by the user. Use only when the user explicitly asks to involve Claude, Anthropic or another model, or invokes $duo.
4 other projects use a version of this.
Adopted by
Most recently taken up by
- yamancan/skillstheir own version, first seen Oct 20, 2026
- agentik-os/claude-code-skillstheir own version, first seen Sep 17, 2026
- heygen-com/hyperframes-community-skillstheir own version, first seen Aug 25, 2026
- juanexpensive/Skillstheir own version, first seen May 25, 2026
Other versions4 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 4
- heygen-com/hyperframes-community-skills★ 155same name
- yamancan/skills★ 2same name
- agentik-os/claude-code-skills★ 1same name
- juanexpensive/Skills★ 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.
The file
We split this skill into the rules it gives an agent
2 rules. 0 are common, 1 is distinctive, and 1 is very specific to this project.
0
Common
Most of the 3 edited versions we read keep these.
1
Distinctive
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
- Distinctive: line 13
If the command still asks for approval, the ChatGPT app has not loaded the rule yet (it needs a restart).
- Very specific: line 37
Read the report.md whose path is printed at the end, then give the user:
“Common” is measured: kept by at least 6 in 10 of the 3 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
duo: Claude x Codex sessions
duo-safe runs every participant as a separate seat with exactly the model and effort given, and records everything in a run folder: exact prompts, replies, raw event streams, tool calls, tokens, credits, and quota before and after. It is the restricted entry point of the duo harness: read-only seats only (no pair mode), no raw Codex config passthrough and no WebFetch.
How to run it (important)
duo-safeneeds network access, so it runs outside your sandbox.~/.codex/rules/duo.rulespre-approves it, but only when it is ONE plain command: no pipes, redirections, heredocs,$(...),&&, orVAR=valueprefixes.- So write any brief longer than one line to a file in the temp folder first, then pass it with
-f <that file>. - If the command still asks for approval, the ChatGPT app has not loaded the rule yet (it needs a restart). Ask the user to approve it.
- Runs take from 30 seconds to many minutes, depending on effort and rounds. Wait for them to finish.
The user's arguments are the spec
The user is an expert who chooses models deliberately. Pass their choices through unchanged: never swap a model, lower an effort, or pick a preset they did not ask for.
- Seat:
engine[:model][@effort][+option...], e.g.claude:opus@max,codex:gpt-6-astra@xhigh+verbosity=high,claude:sonnet@high+web.- Codex options:
verbosity=,summary=,web[=mode],tier=fast. - Claude options:
web,dir=PATH. - Both:
name=,persona=@file,timeout=SEC.
- Codex options:
- Presets:
-p quick|std|deep|max. With no seats and no preset, duo uses the configured default; say which in one line. - Protocol: the first word. If missing, a decision or design is a debate, a diff is a review, an open question for three or more seats is a council, and quick parallel answers are an ask.
duo-safe debate -s SPEC -s SPEC [--rounds N] [--chair SPEC] [--anon] (-C DIR | --no-project) -f /tmp/brief.md
duo-safe review -s SPEC [-s SPEC] --uncommitted -C DIR "focus"
duo-safe council -s SPEC -s SPEC -s SPEC [--chair SPEC] -C DIR -f /tmp/question.md
duo-safe ask -s SPEC [-s SPEC] -C DIR -f /tmp/question.md
duo-safe continue RUN --rounds 1 "moderator note"
duo-safe trace RUN
duo-safe quotaReport back
Read the report.md whose path is printed at the end, then give the user:
- The outcome: converged, NOT converged or stalled, how many rounds, which seats.
- The agreed answer. If the seats did not converge, show the competing final positions side by side. Never present non-convergence as agreement.
- Disputed claims with each side's reason, and any failed citations.
- The cost and the run id.
You operate the harness; you are not a participant. Label any commentary you add as yours. Seats are read-only; you make any changes, and only after the user agrees.
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
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.