rubber-duck
Skill3 starsChanged 3 months agoMITView on GitHub
Rubber-duck a plan, trace, or implementation past a model from a different family and get a second opinion back. Use when a bounded solution is ambiguous, a failure will not surface, the first instance of a new idiom needs confirming, repeated variations keep failing, or a reversible decision would otherwise go to the user. Reach for it whenever another skill calls for a rubber-duck pass; on Copilot CLI it defers to the built-in agent.
22 other projects use a version of this.
Adopted by
Most recently taken up by
- Osipchuk/agent-skillstheir own version, first seen Nov 8, 2026
- jsleemaster/gooblintheir own version, first seen Nov 1, 2026
- vidluther/skillstheir own version, first seen Oct 31, 2026
- damacus/agent-skillstheir own version, first seen Oct 29, 2026
- oneafrikan/the-gridtheir own version, first seen Oct 20, 2026
- kirodotdev/KiroCrewtheir own version, first seen Oct 19, 2026
Used with
Projects with it also use caveman, grilling and 1 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
MicrosoftSkillcaveman
Ultra-compressed response style that reduces output token count while preserving technical accuracy, with intensity levels and auto-clarity safety rules
Also in 3 projects that use this
GoogleSkillgrilling
Grill the user relentlessly about a plan, decision, or idea. Use when the user wants to stress-test their thinking, or uses any 'grill' trigger phrases.
Also in 3 projects that use this
GoogleSkilltdd
Test-driven development. Use when the user wants to build features or fix bugs test-first, mentions "red-green-refactor", or wants integration tests.
Also in 3 projects that use this
Other versions22 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 22
- kirodotdev/KiroCrew★ 4.8ksame name
- thoughtbot/rails-consultant★ 29same name
- jim60105/copilot-prompt★ 22same name
- stevepolitodesign/skills★ 10same name
- The-Artificer-of-Ciphers-LLC/skills-from-the-artificer★ 6same name
- jmcelreavey/devhub★ 5same name
- Osipchuk/agent-skills★ 5same name
- oneafrikan/the-grid★ 4same name
- stormlightlabs/thunderus★ 4same name
- jsleemaster/gooblin★ 3same name
12 more not listed
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 repository29 other files it gives its agents.
- arch-mapSkill
- authenticated-browseSkill
- autonomous-github-actionsSkill
- constraint-challengeSkill
- design-docSkill
- development-loopSkill
- dual-reviewSkill
- explainSkill
- git-history-recoverySkill
- github-api-integrationSkill
- grill-meSkill
- guardrailsSkill
- handoffSkill
- macos-background-app-controlSkill
- macos-photos-librarySkill
- mailboxSkill
- prompt-engineeringSkill
- prototypeSkill
- rotate-sessionSkill
- scoutSkill
- secret-hygieneSkill
- self-compactSkill
- session-transcript-censusSkill
- skill-createSkill
- skill-evaluationSkill
- unattended-runSkill
- visual-proofSkill
- walkthrough-videoSkill
- writing-great-skillsSkill
The file
We split this skill into the rules it gives an agent
2 rules. 0 are common, 2 are distinctive, and 0 are very specific to this project.
0
Common
Most of the 22 edited versions we read keep these.
2
Distinctive
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 33
Read-only is not a formality here.
- Distinctive: line 50
Ask for blind spots, design flaws, and substantive issues.
“Common” is measured: kept by at least 6 in 10 of the 22 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
rubber-duck
Explaining a problem aloud finds the flaw. Explaining it to a different model family finds the flaw your own family is biased to miss — that is the whole delta over talking to yourself, and the reason the listener must not be the model already driving the session.
One pass, advisory only. The duck reads and reasons; it never edits, builds, tests, or lands. You decide what to do with what comes back.
Host
Copilot CLI ships this as a built-in agent, so use that and stop here. The rows below exist because other hosts have no built-in, and the skills in this library call for a rubber-duck pass regardless of where they run.
| Host | Reach a different family by |
| Copilot CLI | the built-in rubber duck agent |
| Claude Code | codex exec -m <gpt-model> --sandbox read-only --skip-git-repo-check |
| Codex CLI | claude -p --model <opus-model> |
Model identifiers move between generations. Resolve the current full,
non-mini, non-codex model at session start rather than trusting an example.
dual-review carries the same transport table; when both run in one session,
resolve the models once.
Read-only is not a formality here. The duck's value is an outside view of work it did not write, and a listener that starts editing the tree stops being a listener. Where the transport offers a read-only sandbox, use it, so the boundary holds mechanically rather than by instruction.
What to send
Send the smallest packet that lets an outsider judge the thing:
- the decision, plan, trace, or diff in question;
- what you are trying to achieve, and what is explicitly out of scope;
- what you already tried and what it did;
- the specific question you want answered.
A duck given the whole repository returns generalities. A duck given one ambiguous decision and its constraints returns an opinion you can act on.
Ask for blind spots, design flaws, and substantive issues. Ask it to say plainly when it would choose differently and why — a duck that only validates is a duck you did not need.
What comes back
Feedback is evidence, not instruction. Take the parts that survive your own check of the code, and say so when you set a point aside; an unexamined "the duck said so" is how a second opinion becomes an unreviewed edit.
One pass is the norm. Reaching for a second round on the same question usually means the packet was unclear rather than the answer, so sharpen the question before spending another pass.
Boundary with dual-review
Reach for the duck to make a decision; reach for dual-review to land a
change.
The duck is one advisory pass on a question still open. dual-review is two
reviewers, adjudication, and a bounded round budget over a change that already
works, and it owns the landing gate. A duck pass never satisfies that gate, and
a change that only needs a second opinion never needs the full protocol.
Skills that offer both name their own threshold — guardrails sets one for
amending an invariant. Follow the calling skill's rule rather than a copy of it
kept here.
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.

