discussion
Skill96 starsChanged 2 days agoMITView on GitHub
Open a structured multi-round discussion that converges - research, explore options, decide together - and persist it as a .scratch work item ending in spec.md and information-ready tickets. For topics too big or contested for a one-shot chat answer.
3 other projects use a version of this.
Adopted by
Most recently taken up by
- Contexera/dsh-agent-teamword for word, first seen Nov 8, 2026
- samchon/nestiatheir own version, first seen Oct 9, 2026
- samchon/typiatheir own version, first seen Oct 8, 2026
Other versions3 projects use a version of this.
Word for word · 1
- Contexera/dsh-agent-team★ 96100% same
Edited · 0
None found.
Wrote their own version · 2
- samchon/typia★ 6ksame name
- samchon/nestia★ 2.2ksame 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 repository12 other files it gives its agents.
- AGENTS.mdAGENTS.md
- docs/AGENTS.mdAGENTS.md
- .scratch/AGENTS.mdAGENTS.md
- ax-initSkill
- ax-reviewSkill
- codebase-ablationSkill
- codebase-designSkill
- diagnosing-bugsSkill
- grillingSkill
- researchSkill
- scratch-reviewSkill
- tddSkill
The file
We split this skill into the rules it gives an agent
8 rules. About 0 are common, 3 are distinctive, and 5 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
5
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 19
Follow .scratch/AGENTS.md: active/<work>/ with its continuation README and a materials/ directory.
- Very specific: line 21
If the repository has no .scratch work area, offer to run ax-init first.
- Very specific: line 31
Keep decision-relevant raw excerpts, worked examples, counterexamples, rejected alternatives, and their provenance in materials/ as the …
- Distinctive: line 43
When the shape is agreed, propose the breakdown — title, blocked-by, and what each ticket delivers — and settle granularity with the user …
- Distinctive: line 45
Follow the repository's generic ticket contract.
- Very specific: line 50
Leave tickets with unfinished blockers or missing decisions, access, references, or verification as pending, with the gap recorded.
- Distinctive: line 56
Update the README's status, last-checked date, frontier, completion conditions, and maintained-document exit as work moves.
- Very specific: line 60
After a coherent ticket set or the work item has been delivered, offer /scratch-review <work-item> for implementation review.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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.
Files
Discussion
A conductor for converging work: a topic that needs multiple rounds — research, exploring alternatives, decisions — and deserves persistence as a .scratch/active/<work>/ work item instead of chat that evaporates at compaction.
The discussion is the interview. Its durable output is a current spec.md, the evidence needed to understand it, and, once the shape is agreed, issues/ tickets. The repository's .scratch/AGENTS.md owns artifact structure, ticket contracts, readiness, delegation, and lifecycle; this skill owns the converging conversation and its handoffs. DESIGN.md is a legacy snapshot name, not a second output.
1. Gate
Discussion-worthy means: multiple rounds needed, real decisions to make, and the outcome will be built or referenced later. A question with a one-pass answer gets answered, not a work item.
2. Open the work item
Follow .scratch/AGENTS.md: active/<work>/ with its continuation README and a materials/ directory. Name the default acceptance owner. Work in the conversation; persist what matters as it crystallizes, not at the end.
If the repository has no .scratch work area, offer to run ax-init first. Without one, keep the discussion in-conversation and say that nothing is being persisted.
3. Research
Pull in facts with the research skill — primary sources, each claim cited, findings saved into materials/. Facts are the agent's job: never ask the user for something that can be looked up.
4. Discuss and preserve evidence
Run the questioning with the grilling discipline: map the topic as a design tree, work it in frontier rounds, numbered questions each carrying a recommended answer; one round, then wait. Settled answers reshape the tree and unlock the next frontier. The questioning is done when the frontier is empty — nothing left silently assumed.
Keep decision-relevant raw excerpts, worked examples, counterexamples, rejected alternatives, and their provenance in materials/ as the discussion moves. Mark whether each material is current evidence or history. Do not wait for the final summary to rescue meaning from a long transcript.
5. Converge
Synthesize — do not transcript, and do not discard the evidence behind the synthesis. spec.md holds the problem in the user's terms, the chosen shape, current decisions with their reasons and source links, explicit boundaries and non-goals, agreed verification seams, and open questions. Move superseded choices out of the current contract or label them as history.
The README routes to the spec and materials and tracks the frontier; it does not become a decision log that a worker must reconstruct. Stable paths, symbols, and decision-dense fragments may be retained with provenance and verification instructions under the repository's scratch rules.
Done means every remaining open question is either non-blocking and explicitly deferred, or prevents the affected tickets from becoming ready.
6. Ticket and establish readiness
When the shape is agreed, propose the breakdown — title, blocked-by, and what each ticket delivers — and settle granularity with the user before writing ticket files.
Follow the repository's generic ticket contract. For each ticket:
- Name one independently reviewable outcome and the evidence suited to its deliverable type.
- Point to exact required/binding sections and separate background reading. Do not assign the whole history or compress away examples that define an abstract requirement.
- State boundaries, settled decisions that must not be reopened, acceptance, and evidence to return.
- Leave tickets with unfinished blockers or missing decisions, access, references, or verification as
pending, with the gap recorded.
Before marking a ticket ready, check both dependency and information readiness. For substantial, ambiguous, or high-impact tickets, recommend /scratch-review <work-item> for design review and use a fresh-context rehearsal: the fresh agent restates outcome, boundaries, required reading, and verification without implementing. A settled-product question means the ticket is not ready.
7. Deliver, review, and close
Update the README's status, last-checked date, frontier, completion conditions, and maintained-document exit as work moves.
Dispatch each ready ticket with the repository root, work-item README, its named reading packet, applicable repository rules/checks, and writable scope. For code, use tdd at the seams already confirmed in the spec; other deliverables use their agreed verification. Workers return changes, observed check results, unverified scope, deviations, and questions rather than claiming Human acceptance.
After a coherent ticket set or the work item has been delivered, offer /scratch-review <work-item> for implementation review; use an earlier per-ticket review only when its stated risk gate requires one. Resolve confirmed findings, then collect any required Human acceptance under validation/.
Closing follows .scratch/AGENTS.md: the acceptance owner completes the tickets, durable conclusions graduate into maintained documents, the archive index is updated, and the work item archives. Do not rewrite older archives to match the new workflow.
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.

