Recipe Design logo

Recipe Design

Community
shinpr
recipe-design

Execute from codebase-scoped analysis through optional ADR decisions to complete Design Doc approval

Overview

Publishershinpr
Repositoryclaude-code-workflows
Skill namerecipe-design
Stars
682
Forks
103
Bundled files
Instructions only
LicenseMIT
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • Self-contained

    Everything the model needs lives in the instructions — no extra files to sync.

  • Open source

    Published by shinpr on GitHub. Read the source before you install it.

Installation

Install the Recipe Design AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/shinpr/claude-code-workflows.git /tmp/claude-code-workflows
mkdir -p .claude/skills
cp -r /tmp/claude-code-workflows/dev-workflows-fullstack/skills/recipe-design .claude/skills/recipe-design
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Recipe Design in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Recipe Design on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Recipe Design is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

Explicit User Instruction: The user explicitly instructs and authorizes every subagent call named in this recipe. Execute each applicable call when its prerequisites are met.

Execute Skill: documentation-criteria before document routing or creation. Execute Skill: llm-friendly-context before writing Agent prompts, handoffs, or generated artifacts. Execute Skill: subagents-orchestration-guide before invoking agents or resolving findings. Before the first finding disposition, read references/review-resolution.md from the loaded subagents-orchestration-guide skill.

Outcome and Ownership

Coordinate the design phase from repository evidence to an approved Design Doc. The user owns product requirements and exclusions; the orchestrator owns convergence readiness, Structural Scale, ADR qualification, evidence selection, and Review Resolution. Named specialists own semantic investigation and artifact authorship.

The Design Doc is always the complete implementation design for Medium/Large work. A qualifying ADR batch narrows technical choices before the Design Doc, which retains the complete flow and implementation boundary.

Requirements: $ARGUMENTS

Flow

text
requirement source -> codebase-analyzer -> scope/decision confirmation [Stop]
                                             |
                               optional ADR batch -> batch review [Stop]
                                             |
                 Design Doc -> code-verifier -> Review Resolution
                                             |
                     document-reviewer -> design-sync -> approval [Stop]

Execute each dependent step after its prerequisite evidence exists. Use Review Resolution for every actionable verifier, reviewer, or design-sync finding. Wait at each [Stop] for explicit user confirmation.

At each Invoke below, build the Agent prompt as a mechanical extraction: copy the named source values into the exact fields, apply only the declared serialization, then invoke immediately.

Step 1: Select the Governing Requirement Source

Use the approved PRD path when one exists. Otherwise use the confirmed requirements verbatim.

Set confirmed_requirement_context to the approved PRD path exactly. Only when no approved PRD exists, use the orchestrator-confirmed convergence record unchanged.

Step 2: Collect Decision Material

Invoke dev-workflows-fullstack:codebase-analyzer:

text
prd_path: [approved PRD path]

or, when no approved PRD exists:

text
requirements: [confirmed requirements verbatim]

Invoke once for the complete confirmed scope. Require one valid JSON result and let the analyzer discover affected paths, responsibility boundaries, and cross-layer contracts. Treat its focus areas as existing-behavior safeguards, not as new requirements.

This independent discovery keeps scope and option convergence grounded in repository evidence rather than the orchestrator's unverified implementation hypothesis.

Step 3: Confirm Scope and ADR Decisions

Execute Skill: requirement-convergence. The orchestrator builds and judges the convergence record from the user request and Step 2 evidence.

Judge all four convergence fields. Assign cost from Step 2 structural evidence and record its unknowns; run the hearing only for fields below ready.

Determine Structural Scale from outcomes and responsibility boundaries. File count is supporting evidence only.

Resolve decisionMaterials.candidateDecisionPoints against the governing requirement source, applicable simplifications, reuse, and invalidations. Remove a point when that evidence already converges on one sufficient approach. For each remaining item, apply documentation-criteria filters in order:

  1. Choice requires judgment between at least two credible, materially distinct options inside confirmed scope.
  2. The selection has durable material impact.

Record every passing item as adrDecisionPoints; an empty list routes directly to the Design Doc. ADR creation is limited to items that pass both filters.

Present the requirement-convergence Scope Confirmation. Place target responsibilities with their strongest file evidence, applicable simplifications with their conditions, and each qualifying ADR decision point with its filter evidence or none under Decision evidence; place Structural Scale with its boundary rationale and the recommended document route under Workflow.

Offer proceed, or correct scope and re-run analysis. Ask a question only when its answer can change a convergence field, the confirmed outcome, or scope. Continue only when every convergence field is ready or weak-but-explicit. [Stop: Scope confirmation].

Step 4: Create and Approve an ADR Batch When Needed

When adrDecisionPoints is non-empty:

  1. Invoke dev-workflows-fullstack:technical-designer once with exact inputs: document_to_create: ADRBatch; confirmed_requirement_context; decision_points as the ordered adrDecisionPoints confirmed in Step 3, unchanged; and decision_materials as the corresponding objects from Step 2 decisionMaterials.candidateDecisionPoints, copied unchanged in that order.
  2. Invoke dev-workflows-fullstack:document-reviewer once with exact inputs: doc_type: ADRBatch, targets: [all returned paths], and confirmed_requirement_context.
  3. Route the reviewer verdict first: pass proceeds with issues: []; needs_revision applies Review Resolution, updates one ADR per path serially, and re-reviews the complete batch; rejected resolves the governing-source conflict before another review.
  4. Present one batch decision only after a pass review. [Stop: ADR batch approval].
  5. After user approval, update each ADR status to Accepted and verify the changed status.

Step 5: Create the Design Doc

Create the complete MVP implementation design from reviewed artifacts and unchanged repository evidence; this keeps the Design Doc traceable to approved sources instead of an orchestrator-authored shadow design.

Invoke dev-workflows-fullstack:technical-designer with exactly:

  • document_to_create: DesignDoc;
  • confirmed_requirement_context;
  • structural_scale;
  • adr_paths: [accepted paths or []];
  • codebase_analysis: [complete Step 2 JSON unchanged].

The Design Doc owns the full end-to-end design and retains all applicable downstream safeguards in the documentation-criteria template.

Step 6: Verify and Resolve Repository Claims

Keep verifier observations unchanged so corrections remain traceable to observed repository evidence instead of becoming orchestrator-authored design instructions.

Invoke dev-workflows-fullstack:code-verifier with doc_type: design-doc and the Design Doc path to verify current premises and feasibility while treating planned behavior as intent.

Apply Review Resolution to every discrepancy before document review. Send only apply findings to a fresh technical-designer update invocation with Operation Mode: update, Existing Document: [Design Doc path], and correction_findings: [complete findings unchanged except for their dispositions]. The designer applies its review-triggered bounded self-verification gate when a finding names an unverified decision-changing premise; this fresh designer is the sole correction specialist and selects the evidence route. Rerun code-verifier after a correction with the previous complete result, dispositions, and correction diff or paths as prior_feedback. Build the single verification_evidence object defined by Review Resolution from the latest result and continue at its convergence condition.

Step 7: Review and Approve

Invoke dev-workflows-fullstack:document-reviewer with exact inputs: doc_type: DesignDoc, target, review_context: creation, the original user requirements verbatim as requirements_verbatim, confirmed_requirement_context, codebase_analysis, and verification_evidence from Step 6.

  • pass: continue.
  • needs_revision: apply Review Resolution, update through a fresh technical-designer invocation using the existing path and complete applied findings, then rerun Steps 6-7 for the affected boundary.
  • rejected: resolve technical governing-source conflicts through Review Resolution; ask the user only when confirmed outcome, desired-future requirements, and non-goals cannot all remain true and the user must choose which changes.

Invoke dev-workflows-fullstack:design-sync for consistency with other Design Docs and apply Review Resolution to actionable conflicts. Report SKIPPED distinctly when only one Design Doc exists.

Present the Design Doc, accepted ADR paths, resolved limitations/declines, and design-sync result. [Stop: Design approval].

Completion Criteria

  • Scope and Structural Scale were confirmed from outcomes and responsibility boundaries.
  • ADRs exist only for decision points passing both filters, and the complete batch received one review and approval.
  • A Design Doc exists regardless of whether ADRs were needed.
  • Applicable existing-behavior, contract, assumption, equivalence, and verification safeguards reached the Design Doc.
  • Review Resolution routed only needs_revision issues into correction work.
  • All stop points received explicit user confirmation.

Frequently asked questions

What does the Recipe Design AI skill do?

Execute from codebase-scoped analysis through optional ADR decisions to complete Design Doc approval

Why use Recipe Design on TypingMind?

Because you install it once and use it with any model. Recipe Design is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Recipe Design in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/shinpr/claude-code-workflows/tree/main/dev-workflows-fullstack/skills/recipe-design. TypingMind reads its SKILL.md and installs it as a skill you can enable per chat.

Which AI models can use Recipe Design?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Recipe Design?

As many as you like. As long as a model supports skills, you can use Recipe Design with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Recipe Design AI skill free?

Yes. It is published on GitHub by shinpr under the MIT license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇