spec-design
Skill1 starsChanged 3 months agoMIT-0View on GitHub
Turn a specs/<name>.yaml spec into an AI-DLC design-doc.md, entering the pipeline at the design phase (spec is already-refined requirements, so story/refine are skipped)
8 other projects use a version of this.
Adopted by
Most recently taken up by
- xsyubin/clue-lenstheir own version, first seen Nov 15, 2026
- warrendeanlangeveldt/code-kittheir own version, first seen Nov 15, 2026
- komluk/scaffoldingtheir own version, first seen Nov 15, 2026
- cldanielhernandez/claude-skillstheir own version, first seen Aug 8, 2026
- fernandoxavier02/Pipeline-Orchestratortheir own version, first seen Jun 29, 2026
- yisean/claude-spec-skillstheir own version, first seen Jun 15, 2026
Other versions8 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 8
- MagicCube/agentara★ 503same name
- komluk/scaffolding★ 16same name
- alanvaa06/forge-master★ 2same name
- warrendeanlangeveldt/code-kit★ 2same name
- fernandoxavier02/Pipeline-Orchestrator★ 1same name
- cldanielhernandez/claude-skills★ 0same name
- xsyubin/clue-lens★ 0same name
- yisean/claude-spec-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.
- CLAUDE.mdCLAUDE.md
- ai-dlcSkill
- commitSkill
- design-docSkill
- pr-draftSkill
- story-newSkill
- story-refineSkill
- tasksSkill
The file
We split this skill into the rules it gives an agent
1 rule. 0 are common, 1 is distinctive, and 0 are very specific to this project.
0
Common
Most of the 6 edited versions we read keep these.
1
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 49
Every task's Acceptance must cover at least one DoD checkbox.
“Common” is measured: kept by at least 6 in 10 of the 6 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
/spec-design — spec → design doc (AI-DLC bridge)
Bridges a factory spec (specs/<name>.yaml) into the AI-DLC inception pipeline. A spec is not a rough idea — it is already-refined requirements (tools, inputs, outputs, errors, dependencies, deploy target all resolved). So this skill enters AI-DLC at phase 3 (design), skipping story-new and story-refine, and produces a design-doc.md whose task map the existing /tasks skill then expands. Building is manual from those tasks.
The spec is the source of truth: this skill is re-runnable, and the design doc points back at the spec (not a story.md) as its authority. When the spec changes, re-run this skill to regenerate the design doc, then /tasks.
Pipeline position: specs/<name>.yaml → /spec-design → design-doc.md → /tasks → tasks/ → (manual build) → /pr-draft.
Input
The spec name, e.g. /spec-design pubmed — reads specs/pubmed.yaml. Optional slug for the output folder (defaults to the spec name).
Steps
- Read the spec
specs/<name>.yaml. If it is missing, STOP and list the specs underspecs/. Parse its five layers:server,dependencies,types,tools, and each tool's inputs/constraints/output/errors/behavior. Also readspecs/SCHEMA-NOTES.mdif present — it records how spec fields map to code (e.g. a reserved word likeabstract→abstract_text,clampin core logic vs.rejectat the input boundary). - Resolve the target folder
.local/user-stories/<slug>/. If adesign-doc.mdalready exists, say so and confirm regeneration (the spec is source of truth, so regeneration is expected — but never clobber silently). - Read the template in
references/design-template.md(bundled — a copy of the/design-doctemplate so output is identical in shape to a hand-run design phase). Match its section order exactly, so/tasksconsumes it unchanged. - Derive each section from the spec (see the mapping below). Do not re-elicit requirements — translate what the spec already states into structure. Verify only the how (package versions, SDK API shapes) that the spec deliberately leaves to the engine; note the verification date. Do not invent versions from memory.
- Show the draft design doc — especially the architecture and the task map — and confirm before writing. The task map is the contract
/tasksexpands, so it is the key gate. - On confirmation, write
.local/user-stories/<slug>/design-doc.mdand suggest/tasks <slug>.
Spec → design-doc mapping
Derive, don't elicit. Each design-doc section comes from spec fields:
| design-doc section | Comes from |
| Goal & non-goals | server.description; non-goals from dependencies/scope the spec omits, plus recorded ADR decisions (Runtime-only, stateless default). |
| Architecture | The invariant spine (the FastAPI app + AgentCore container contract: entrypoint, ARM64 Dockerfile, shared adapters from src/agent-core/shared/) + a variant core per the spec's tools/types/dependencies. Reuse the three-layer split: pure/IO core (no framework types) → tool/agent adapter → FastAPI transport spine. Draw the file tree from the tools and shared types. |
| Key dependencies | The spine set (strands-agents, fastapi/uvicorn, pydantic, AWS Lambda Powertools) + whatever dependencies.outbound_http implies (an HTTP client such as httpx for HTTP deps, a JSON/XML parser per response_format). State capabilities → concrete packages here; the spec stays capability-level. |
| Task map | The regular decomposition: T1 scaffold (package layout + pyproject/deps), T2 core (Pydantic types + per-tool logic, unit tests), T3 tool/agent adapter (a @tool per spec tool + input models), T4 FastAPI transport spine, T5 integration test, T6 ARM64 Dockerfile, T7 README. Add a dependency/config task when dependencies is non-trivial (HTTP client + rate limit + response parsing), and capture-fixtures guidance when response_format is a structured format. |
| Cross-cutting contract | The AgentCore Runtime invariant (ARM64, FastAPI on 0.0.0.0:8080, /ws + /invocations + /ping, stateless container per server.stateful, auth per server.auth) — restate and map each row to the task that enforces it. |
| Decisions & open questions | Decided items from the spec + linked ADRs (target, stateful, auth). Constraint modes (clamp/reject), nullable outputs, and computed fields become explicit decisions with the task that owns them. Flag anything the spec leaves genuinely open. |
Definition of Done (in place of a story)
A spec has no story.md, so derive the DoD — the observable, testable outcomes every task maps to — directly from the spec and state it near the top of the design doc:
- One checkbox per
tool: it appears intools/listwith the spec's input schema, and a representative call returns the spec's output shape. - One checkbox per tool
errorcase: the stated condition yields the stated model-facing message (surfaced as a tool error, per the framework's convention — not a success envelope). - Constraints honored:
clampbounds applied in the core;rejectinputs error before any side effect. - The cross-cutting contract holds (builds ARM64; FastAPI serves
0.0.0.0:8080with/ws+/invocations+/ping; stateless/stateful per spec).
Every task's Acceptance must cover at least one DoD checkbox; flag any gap rather than papering over it.
Principles
- Spec is source of truth. The design doc references
specs/<name>.yamlas its authority (a link at the top, in place of the usualstory.mdcompanion link). Re-running this skill after a spec change regenerates the doc. - Structure here, API surface in
/tasks, implementation in the manual build. Same rule as/design-doc: no function bodies in the design doc — only architecture, dependencies, and the task map. - Honor recorded decisions. The spine is fixed by ADRs (where present); don't re-litigate CDK-vs-Terraform, Runtime-vs-Gateway, or stateless-default — reference the ADRs.
- Thin over
/design-doc. This skill is the design phase for spec-driven work; it deliberately produces the same artifact shape so everything downstream (/tasks, build,/pr-draft) is unchanged.
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.

