spec-drafter
Skill0 starsChanged 5 months agoMITView on GitHub
Converts agreed intent and research into a structured spec (the what & why) — objective, requirements with acceptance criteria, architecture, ADRs, and Always/Ask-First/Never boundaries. Use after intake and research, for Medium+ work. Produces spec.md.
1 other project uses a version of this.
Adopted by
Most recently taken up by
- 10CG/aria-plugintheir own version, first seen Oct 20, 2026
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- 10CG/aria-plugin★ 1same 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.
- .cursor/rules/000-harness-bootstrap.mdcCursor rule
- .cursor/rules/001-planning-safety.mdcCursor rule
- .cursor/rules/002-source-discipline.mdcCursor rule
- .cursor/rules/003-anti-rationalization.mdcCursor rule
- .cursor/rules/004-memory-and-resume.mdcCursor rule
- doubt-researchSkill
- intake-interviewSkill
- master-plannerSkill
- plan-grillerSkill
- project-memorySkill
- security-rollbackSkill
- task-decomposerSkill
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 0 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 22
Write .cursor/plans/<project>/spec.md.
- Very specific: line 42
Do not write the task graph yourself — that is task-decomposer's job.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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
Spec Drafter
You produce the shared source of truth between the user and the implementer. Code without a spec is guessing; the spec's whole value is forcing clarity before code.
Integrates the six-area spec + Always/Ask-First/Never boundaries + "reframe requirements as success criteria" model from addyosmani/agent-skills spec-driven-development (MIT), the conversation→PRD synthesis idea from mattpocock/skills to-prd (MIT), and the numbered-artifact discipline from anton-abyzov/specweave (MIT). See docs/ATTRIBUTION.md.
Step 1 — Surface assumptions, then draft spec.md
Lead with the assumptions block (carried over from intake) so misunderstandings surface before content:
ASSUMPTIONS I'M MAKING:
1. ...
-> Correct me now, or I proceed with these.Write .cursor/plans/<project>/spec.md. Include only sections that carry weight — do not pad with empty headings.
Core sections:
- Objective — what we're building and why; who the user is; what success looks like.
- Non-Goals — what this explicitly does not do.
- Requirements — each with verifiable acceptance criteria (a checkbox you could turn into a test). Reframe vague asks as concrete targets ("faster" → "LCP < 2.5s on 4G").
- Architecture — system boundaries, data model, API/interface contracts, technology stack with one-line justifications.
- Boundaries — the three-tier set: Always do / Ask first / Never do.
Conditional sections (include when relevant):
- Research Findings — verified facts with source links (from
doubt-research). Unverifiable items marked[NO RELIABLE DATA]. - Dependencies — table of package · version · license · maintained? · justification.
- Security Considerations & Deployment & Rollback — required for High-Risk / Large (delegated to
security-rollback). - ADRs — for each significant "why X over Y" decision, using
../master-planner/references/ADR-FORMAT.md(Context → Decision → Consequences). Offer an ADR only when the decision is hard to reverse, surprising without context, AND a real trade-off. - Open Questions — unresolved
[NEEDS CLARIFICATION]markers.
The .cursor/plans/_template/spec.md template mirrors this structure.
Step 2 — Hand off to decomposition
Do not write the task graph yourself — that is task-decomposer's job. Pass it the validated spec. Keeping spec (what/why) separate from tasks (how) keeps each reviewable on its own.
Step 3 — Update the manifest
Set Phase: SPEC_COMPLETE (or SPEC_COMPLETE_NEEDS_GRILL for Large/High-Risk). Record artifact paths and a resume instruction.
Step 4 — Keep the spec alive
The spec is a living document: update it when a decision or scope changes (update the spec first, then implement), commit it alongside code, and reference its sections from PRs.
Self-check before handoff
- Every requirement has verifiable acceptance criteria.
- Architecture states system boundaries and data flow.
- Dependencies list version, license, and maintenance status.
- Boundaries (Always/Ask-First/Never) are present.
- No unresolved
[NEEDS CLARIFICATION]left unflagged to the user. - For High-Risk: security + rollback sections are present.
Anti-Rationalization
| Excuse | Reality |
| "Acceptance criteria are implicit" | Implicit means untestable. Write them out. |
| "I'll write the spec after coding" | That's documentation, not specification. The value is clarity before code. |
| "This dependency is fine, everyone uses it" | Record version, license, and maintenance status anyway. |
| "The architecture is simple enough to skip" | Simple architectures document fast. Write the boundaries and the ADR. |
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.