create-rfc
Skill819 starsChanged 18 days agoApache-2.0View on GitHub
Research, interview, and develop an engineering RFC from an idea, notes, or an existing draft. Use when asked to create or substantially develop an RFC; ordinary questions and small document edits do not need this workflow.
5 other projects use a version of this.
Adopted by
Most recently taken up by
- NVIDIA/OpenShelltheir own version, first seen Nov 12, 2026
- Ronnasayd/aipptheir own version, first seen Oct 20, 2026
- emiliosheinz/agent-skillstheir own version, first seen Oct 17, 2026
- mekenthompson/ProductOStheir own version, first seen Jul 23, 2026
- popovicsandras/agent-skillstheir own version, first seen Jul 6, 2026
Other versions5 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 5
- NVIDIA/OpenShell★ 22ksame name
- emiliosheinz/agent-skills★ 20same name
- mekenthompson/ProductOS★ 1same name
- Ronnasayd/aipp★ 1same name
- popovicsandras/agent-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.
In this repository8 other files it gives its agents.
- AGENTS.mdAGENTS.md
- contributeSkill
- generate-slideSkill
- propose-featureSkill
- releaseSkill
- review-prSkill
- review-rfcSkill
- update-rfcSkill
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 14
Read the supplied material, relevant repository guidance, and existing RFC conventions.
- Distinctive: line 40
Use Mermaid when a diagram explains a relationship more clearly than prose:
“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
Create RFC
Produce one self-contained RFC that engineers outside the conversation can review. Research establishes facts; the interview resolves consequential decisions. Maintain the same document throughout, using the Rust-based RFC template. This skill ends with RFC review, not implementation or publication.
This is a public repository. Use public evidence or independently developed material suitable for public review. Keep restricted research outside the checkout; removing links or attribution does not make internal findings publishable. See RFC conventions for location and conventions.
Establish the starting point
Read the supplied material, relevant repository guidance, and existing RFC conventions. Identify the problem, audience, constraints, settled decisions, and consequential gaps. An existing draft is the starting artifact: preserve its accepted decisions and focus on contradictions or missing reasoning.
In this repository, use docs/rfcs/NNN-title.md and record status near the title. Select the next available number starting at 001. Outside a repository, use a suitable user-visible working directory. Keep subsequent revisions in the same file. If the destination is shared, prepare a private copy until publication is authorized.
Create an initial RFC with established facts and clearly marked open questions. Use it as the single design document rather than creating separate glossaries, context documents or research reports. Place terminology near the end, followed by a references section when needed. Required private execution notes may hold resumption state, but must not become an alternative design source. Keep incidental experiment artifacts out of the RFC's narrative.
Research and interview in rounds
Map the design as decisions and their dependencies. In each round, ask the independent consequential questions whose prerequisites are settled; defer questions that depend on unanswered ones. Use manageable batches, stable question numbers, concrete scenarios, and a recommended answer with its main tradeoff. Use the host's question tool when suitable, otherwise chat. Allow the user to propose a different answer.
Investigate factual questions yourself. Read relevant code, primary documentation, and available connected sources before asking the user to resolve uncertainty. Use bounded independent research agents when available and useful; otherwise perform the research directly. Keep synthesis and decisions with the lead. Do not delegate a product decision to a research agent.
Small, reversible local experiments may resolve a design question. State what the experiment tests, preserve existing work and data, and stop once the evidence answers it. Ask before substantial environment setup or implementation; keep external mutations within explicit authorization. An RFC request alone does not authorize deploying a prototype, changing shared systems, or messaging others.
Distinguish source-supported capability, observed behavior, proposed design, and unverified integration. Evidence from one build or fixture establishes only what was exercised. Keep research provenance in working context or required private notes. In the RFC, cite only sources that help a reviewer assess a material claim, following the reference policy below. Disclose a relevant coverage gap rather than converting it into a universal limitation.
After each answer, update the RFC immediately and recompute the open decisions. Reflect user corrections throughout prose, diagrams and examples. Preserve the current design rather than appending an interview transcript. Do not reopen settled choices without new evidence or a contradiction; explain that evidence when reconsideration is needed.
Grill behavior, scope, interfaces and meaningful tradeoffs. Choose routine, reversible implementation details with engineering judgment. Leave technical checks that require implementation in unresolved questions, with the concrete evidence needed to resolve them. Exhaustive questioning about incidental details is not the objective.
Write for fellow engineers
Read the template when drafting or restructuring. Lead with the proposal and its effect, then explain the problem and teach the user-visible behavior through a concrete scenario before introducing implementation detail. Use direct, familiar language and connected paragraphs. Include technical detail where it helps assess a contract, tradeoff or failure mode; define domain-specific terms in the terminology section. Label illustrative schema fields or routes as provisional until selected.
For repository RFCs, keep reader-facing references within that repository unless the user requests external references. External documents, meetings and issue trackers may inform research; incorporate their relevant decisions and reasoning so readers can understand the proposal without opening them. Use descriptive links to repository files or pinned source revisions near material claims. Keep raw hashes, local checkout paths and source inventories in working notes rather than the prose. Apply the user's reference policy when working outside a repository.
Use Mermaid when a diagram explains a relationship more clearly than prose:
- Component views for architecture, with subgraphs for meaningful ownership or deployment boundaries. Collapse platforms whose internals are irrelevant to the proposal.
- Sequence diagrams for request, callback and background flows; show where handlers execute and where persistence happens.
- State diagrams for lifecycle transitions and terminal states.
Include code, resource or API examples where they clarify the contract. Keep examples consistent with permissions, identity, lifecycle and error semantics. Do not invent an existing framework or component merely to label a diagram. Use tables for genuinely comparable concepts, not as a substitute for explanatory prose.
Write a snapshot of the proposed design. Give each rule or rationale one clear home; elsewhere use only the short restatement needed to understand a flow. Retain alternatives only when they explain a real engineering tradeoff. Remove abandoned implementation commentary, conversation history, repeated defensive contrasts, and task-management restrictions from the reader-facing RFC. Localize material limitations and distinguish accepted behavior from implementation questions and future work. Combine or omit template sections when that improves the reading order; scale length to the decisions reviewers must assess.
Complete and review
The interview is complete when consequential decisions are resolved or explicitly deferred with the user's agreement, and each unresolved question says what would resolve it. Do not treat silence as agreement.
Perform a final editorial pass from the perspective of an engineer who has not seen the conversation. Check that the proposal and rationale stand on their own, references follow the chosen policy, and repeated evidence or caveats have been consolidated. Compare the result against accepted requirements, exact interfaces, diagrams, examples and terminology so shortening preserves the design. Verify local links and example syntax where practical; validate or render diagrams when tooling is available. State any material validation limitation without implying a check ran. Use an independent bounded review when it adds meaningful confidence.
Present the complete RFC for the user's review, naming any remaining decisions or technical gaps. Ask whether it captures the shared understanding; revise the same artifact until it does. Do not require another approval for each routine draft edit. Implementation, commit/push, PR creation and shared publication are separate actions requiring their own authorization; honor authorization already given.
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.

