sdd-yolo
SDD fast path. Runs the full pipeline (spec, plan, analyze, implement and converge until done, review, archive) with a single confirmation gate before implementation begins. Stops automatically if analyze says FIX-FIRST, converge cannot close, or review finds Critical or Major issues. Use when you want to ship a well-understood feature with minimal interruptions.
- License 许可
- MIT
- 最近更新
- 2026-11-14
- Stars
- ★ 18
- Forks
- 6
01 前言 Front Matter
- name
- sdd-yolo
- description
- SDD fast path. Runs the full pipeline (spec, plan, analyze, implement and converge until done, review, archive) with a single confirmation gate before implementation begins. Stops automatically if analyze says FIX-FIRST, converge cannot close, or review finds Critical or Major issues. Use when you want to ship a well-understood feature with minimal interruptions.
- argument-hint
- <feature description>
02 资源文件 1
- 参考文档skill.json472 B
03 SKILL.md 正文
SDD: YOLO Full Pipeline
You are running the full SDD workflow end-to-end with minimal interruptions. The pipeline is: spec → plan → analyze → implement ⇄ converge → review → archive.
There is exactly one confirmation gate: after the spec and plan are produced, before implementation begins. Everything else runs automatically.
Required Inputs
Before starting, collect these inputs. If any are missing, ask for them now: do not proceed without them.
| Input | Description | Example |
|---|---|---|
feature_description |
The feature to build end-to-end | "Add JWT authentication with refresh token support" |
Steps
Step 0: Validate Inputs (ALWAYS DO THIS FIRST)
Check the conversation for feature_description and for docs/project.md.
- If
docs/project.mddoes not exist → stop and tell the user to run/sdd-initfirst. - If
feature_descriptionis present → proceed to Phase 1. - If
feature_descriptionis missing → ask: "What feature would you like to build?" Do NOT proceed until the user provides it.
Phase 1: Spec
Follow the full sdd-feature process:
- Read
docs/project.md. - Analyse the request:
feature_description(collected in Step 0). - Create the spec directory
docs/specs/<yyyymmddHHMM>-<feature-name>/and writefeature.mdin it, in the ears-1 format ofsdd-feature:Spec-Format: ears-1header, Glossary, EARS acceptance criteria. Anything you cannot decide becomes an inline[NEEDS CLARIFICATION: ...]marker (at most 3). - If markers remain, resolve them with the user one question at a time, each with a
recommended answer, as
sdd-refinedoes. The pipeline does not continue with open markers. - Run
python3 ~/.claude/scripts/spec_lint.py --ready <spec-directory>and fix every finding. - Print a compact summary of the spec (3 to 5 bullet points, not the full file).
Phase 2: Plan
Follow the full sdd-plan process immediately after Phase 1:
- Read
feature.mdanddocs/project.md. - Write
plan.mdin the task grammar- [ ] T## [AC-##] <path>: <action>, at most 20 tasks. - Run the linter again: every criterion must be covered by a task.
- Run
/sdd-analyzeas an independent verifier subagent. OnFIX-FIRST, fix the spec or the plan and re-run it; do not reach the gate with aFIX-FIRSTverdict. - Print a compact summary of the plan (task names only, not full detail) and the analyze verdict.
Confirmation Gate
Present the following prompt and wait for the user's response before continuing:
## Ready to implement
Spec: feature.md ✓
Plan: plan.md ✓
[Compact spec summary: 3–5 bullets]
[Plan steps: numbered list of step names]
Type PROCEED to start implementation, or describe any changes you want first.
- If the user types PROCEED (or equivalent confirmation): continue to Phase 3.
- If the user requests changes: apply them to
feature.mdand/orplan.md, show what changed, then re-present the gate. - If the user aborts: stop and leave
feature.mdandplan.mdin place for manual continuation.
Phase 3: Implement
Follow the full sdd-implement process:
- Read
plan.md,feature.md, anddocs/project.md. - Execute each step in
plan.mdin order. - Compile and run tests after each layer. Fix failures before moving on: never carry failures forward.
- Do not introduce new dependencies without flagging them to the user.
- After all tasks, run the full test suite once and print the completion summary (files created and modified, test result). Do not grade the criteria yourself.
Converge loop
- Run
/sdd-convergeas an independent verifier subagent. - On
GAPS, implement the tasks it appended under## Convergence <n>, then run it again. - Stop the pipeline if the same criterion stays unmet after 3 converge passes. Continue only
on
CONVERGED. When you stop, print:
## YOLO Pipeline Stopped: converge did not close
Feature: <Feature Name>
Unmet after 3 passes: <AC ids, with the converge evidence for each>
feature.md and plan.md stay in place. Decide whether the criterion or the code is wrong:
/sdd-refine for the criterion, /sdd-implement for the code, then /sdd-converge.
Phase 4: Review
Follow the full sdd-review process immediately after Phase 3:
- Run
git diff main...HEAD --name-onlyto determine changed files. - Review across the 7 quality dimensions (language and framework practices, security, duplication, design, performance, test quality, observability), quoting the converge verdict.
- Produce the full structured review report.
Phase 5: Archive or Stop
Evaluate the review verdict:
If verdict is ✅ Ready to merge OR 🟡 Merge after minor fixes
Proceed automatically to archive:
- Follow the full
sdd-archiveprocess. - Update
docs/project.md(features list, architecture decisions, API surface, env config). - Show the proposed
project.mdchanges and ask for confirmation before writing. - The spec directory
docs/specs/<yyyymmddHHMM>-<feature-name>/stays where it is. - Create
README.mdin it.
Then print the final pipeline summary:
## YOLO Pipeline Complete ✓
Feature: <Feature Name>
Spec directory: docs/specs/<yyyymmddHHMM>-<feature-name>/
Phase results:
Spec ✓
Plan ✓ (analyze: READY)
Implement ✓ (N files created, M files modified)
Converge ✓ (CONVERGED after <k> pass(es))
Review ✓ (<verdict>)
Archive ✓
Next: commit the spec directory and docs/project.md to version control.
If verdict is 🟠 Requires fixes and re-review OR 🔴 Do not merge
Stop. Do not archive.
Print:
## YOLO Pipeline Stopped: Review issues require attention
Feature: <Feature Name>
Phase results:
Spec ✓
Plan ✓
Implement ✓
Converge ✓
Review ✗ (<verdict>)
Archive skipped
Critical/Major findings must be resolved before archiving.
Fix the issues above, then run /sdd-review to re-review, and /sdd-archive when clean.
Leave the spec directory as it is so the user can continue manually.
04 同仓库更多 Skills
yeet
Use only when the user explicitly asks to stage, commit, push, and open a GitHub pull request in one flow using the GitHub CLI (`gh`).
wrangler
Cloudflare Workers CLI for deploying, developing, and managing Workers, KV, R2, D1, Vectorize, Hyperdrive, Workers AI, Containers, Queues, Workflows, Pipelines, and Secrets Store. Load before running wrangler commands to ensure correct syntax and best practices. Biases towards retrieval from Cloudflare docs over pre-trained knowledge.
workspace-skill-discovery
Discover and include ALL skills across the workspace — not just global ones. Use at the start of every session or when the user asks about available skills. Ensures project-level skills (under .claude/skills/ or skills/ in any repo) are never missed.
workers-best-practices
Reviews and authors Cloudflare Workers code against production best practices. Load when writing new Workers, reviewing Worker code, configuring wrangler.jsonc, or checking for common Workers anti-patterns (streaming, floating promises, global state, secrets, bindings, observability). Biases towards retrieval from Cloudflare docs over pre-trained knowledge.
winui-app
Create, build and design WinUI 3 desktop apps with C# and the Windows App SDK following Microsoft's official guidance and CommunityToolkit. Covers XAML, navigation, theming, accessibility and deployment.
windows-brain-script-portability
Scripts in ~/.claude/scripts/ that run on a Windows brain: use the python3.cmd shim and not bare python3, and reconfigure stdout to UTF-8 if you print symbols. Use it when writing a brain script or triaging a Windows-only crash.