readback
Skill0 starsChanged 42 days agoMITView on GitHub
Proves the request was understood before anyone builds anything — states the outcome rather than paraphrasing the words, marks every gap it had to fill as said, inferred or guessed, names the boundary, argues the strongest rival reading and lists where you would catch a misunderstanding first. Builds nothing, edits nothing, plans nothing. Use when the user says "/readback", "did you understand me", "what did you understand", "tell me back what I asked", "repeat it back", "say it in your own words first", "before you start, tell me what you think I want", or the Turkish equivalents "beni anladin mi", "ne anladin", "anladigini soyle", "baslamadan once ne anladigini yaz". For a fixed rule pass that fixes what it finds use lint, for redesigning an interface use compose.
5 other projects use a version of this.
Adopted by
Most recently taken up by
- iurysza/agent-skillstheir own version, first seen Nov 1, 2026
- menot-you/n-soultheir own version, first seen Nov 1, 2026
- agentydragon/ducktapetheir own version, first seen Oct 9, 2026
- wilbeibi/wilbeibi-skillstheir own version, first seen Sep 7, 2026
- backnotprop/brotheir own version, first seen Jul 12, 2026
Other versions5 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 5
- backnotprop/bro★ 176same name
- agentydragon/ducktape★ 11same name
- iurysza/agent-skills★ 5same name
- wilbeibi/wilbeibi-skills★ 2same name
- menot-you/n-soul★ 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, 2 are distinctive, and 0 are very specific to this project.
0
Common
Most of the 5 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 12
Builds nothing (MUST): No file written, no code, no approach, no step list, no work begun.
- Distinctive: line 24
Ends without a plan (MUST): No steps, no schedule, no *shall I begin*.
“Common” is measured: kept by at least 6 in 10 of the 5 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
readback — prove it landed, before anyone builds anything
Hand the request back as consequences the reader can reject line by line — never as a paraphrase.
Rules
- Builds nothing (MUST). No file written, no code, no approach, no step list, no work begun. Source is read only far enough to make nouns concrete — the real file, function, column — never far enough to solve.
- Every line is rejectable and adds something the request only implied (MUST) — an outcome, a decision, a boundary, a consequence. The user's own words appear only as evidence, never as the claim. The closing correction line is the one exemption.
- Every claim is marked (MUST): said — traceable to a phrase in the request; inferred — follows from the context, the repo or an earlier turn; guessed — picked with nothing behind it. A guess stays a guess however reasonable. When stacked guesses would let two honest readings produce different work, say the request is underspecified.
- The rival reading is argued at its strongest (MUST), followed by the one question that separates it from yours. Where the request admits one reading, say so; never invent a rival.
- Ends without a plan (MUST). No steps, no schedule, no shall I begin.
- Shorter than what it checks, one screen at most — except a request too short to hold every section, where each section collapses to one line instead. Written in the request's language, headings included. No acknowledgement; open on the first claim. A question appears only if its answer changes the work, and carries what changes.
Target: the instruction just given, or what the argument names — a file, a ticket, a paragraph. "The whole conversation" means the standing instructions, earlier corrections included.
Output
These headings in this order, each dropped when it has nothing to say:
The job — one sentence: the outcome, not the action. If it cannot be made concrete, say understanding failed.
Done looks like — one to three lines: what exists, behaves or reads differently once finished.
| What the request left open | My reading | Level | If it is the other way |
Highest consequence first. Past roughly seven forks, stop and say the request is underspecified.
Not included — what a reasonable reader might expect here and will not get.
Rival reading — the strongest alternative, then the question that separates it from yours.
Where you would catch me — one to three concrete observables where a wrong reading shows first.
Needs an answer — only questions that change the work, each with what changes.
Close with one line inviting a correction, and nothing after it.
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.

