mcpbeat

Recipe Design

shinpr/recipe-design

Execute from codebase analysis to design document creation

3k tokens
context cost
the whole folder, loaded on every use
1
files
instructions only
0
copies elsewhere
how many repositories repackaged it
664
stars on the repo
on the repository, not the skill itself

Install

one command, takes just this skill from the repository
npx skills add https://github.com/shinpr/claude-code-workflows --skill recipe-design

The instruction itself

10 sections, as written by the author

Execute Skill: llm-friendly-context before writing Agent prompts, handoffs, or generated artifacts.

Execute Skill: subagents-orchestration-guide before making workflow decisions, invoking agents, or resolving findings.

Context: Dedicated to the design phase.

Orchestrator Definition

Core Identity: "I am an orchestrator." (see subagents-orchestration-guide skill)

Local authority gate: Make this recipe's workflow decisions and validate each returned result directly; delegate semantic deliverable production to the named specialist. The scope bootstrap locates seed files; the named specialists own semantic investigation and artifact authorship.

Review Resolution Gate [MANDATORY]: Resolve every actionable deliverable-review finding through subagents-orchestration-guide Review Resolution before correction or progression; include declined IDs with governing reasons and evidence in the final user report.

Before the first finding disposition, read references/review-resolution.md from the loaded subagents-orchestration-guide skill.

Execution Protocol:

  • Invoke named specialists for deliverable production — pass data between them and validate their results. Step 1 is a recipe-local read-only scope bootstrap limited to locating seed files.
  • Run the design flow below in order:
  • Execute: scope bootstrap → codebase-analyzer → [Stop: Scope confirmation] → technical-designer → code-verifier → document-reviewer → design-sync
  • code-verifier and design-sync apply when the design output is a Design Doc; both are skipped for ADR-only
  • Stop at every [Stop: ...] marker → Wait for user approval before proceeding
  • Scope: Complete when design documents receive approval

subagents-orchestration-guide usage: Use the guide for orchestration principles (Delegation Boundary, Decision precedence, Execution Boundary), the Scale Determination table, and handoff contracts HC-02 onward. This recipe's start order and subagent prompts supersede the guide's requirement-analyzer-origin flow, First Action Rule, HC-01, and Agent-Specific Prompt Content.

CRITICAL: Execute document-reviewer, design-sync (for Design Docs), and all stopping points — each serves as a quality gate. Skipping any step risks undetected inconsistencies.

Workflow Overview

Requirements → scope bootstrap → codebase-analyzer → [Stop: Scope confirmation]
                                                            ↓
                                                    technical-designer
                                                            ↓
                                                    code-verifier → document-reviewer
                                                            ↓
                                                       design-sync → [Stop: Design approval]

Scope Boundaries

Included in this skill:

  • Scope bootstrap: locating seed files so codebase-analyzer receives a populated input
  • Codebase analysis with codebase-analyzer (entry point of the design phase)
  • Scope confirmation with the user, grounded in codebase-analyzer findings
  • ADR creation (if architecture changes, new technology, or data flow changes)
  • Design Doc creation with technical-designer
  • Design Doc verification with code-verifier (before document review)
  • Document review with document-reviewer
  • Design Doc consistency verification with design-sync

Responsibility Boundary: This skill completes with design document (ADR/Design Doc) approval. Work planning and beyond are outside scope.

Execution Flow

Requirements: $ARGUMENTS

Step 1: Scope Bootstrap

codebase-analyzer requires a populated requirement_analysis.affectedFiles. Build that seed with a lightweight, orchestrator-local pass — locating files only, with no deep reading and no design decisions:

  • Extract candidate keywords from the user requirements (feature names, domain nouns, identifiers).
  • Search the repository with Bash (rg, or grep when rg is unavailable) for files matching those keywords.
  • Collect the matched file paths as the seed affectedFiles.
  • When the search returns no files: ask the user which files or modules the design targets (AskUserQuestion), and use that answer as affectedFiles before invoking codebase-analyzer. If the user confirms no related code exists, report that codebase-grounded design does not apply and confirm with the user how to proceed.
  • When the search returns more than ~20 files: the keywords are too broad for a focused design scope. Present the most relevant candidates to the user (AskUserQuestion) and confirm the seed affectedFiles before invoking codebase-analyzer.

This step locates seed files only. Reading files in full, tracing dependencies, and analysis remain codebase-analyzer's responsibility.

Step 2: Codebase Analysis

Invoke codebase-analyzer with its existing schema. The orchestrator constructs requirement_analysis from the Step 1 seed.

  • Invoke codebase-analyzer using Agent tool
  • subagent_type: "dev-workflows-fullstack:codebase-analyzer", description: "Codebase analysis"
  • prompt: include
  • requirements: the user requirements verbatim
  • requirement_analysis: a JSON object with all four fields — affectedFiles (Step 1 seed), purpose (the user requirements), scale (provisional value from the Scale Determination table applied to the seed file count), technicalConsiderations ({ constraints: [], risks: [], dependencies: [] } — the bootstrap performs no analysis, so the object is present with empty lists)
  • Expected action: analyze the seed files and produce design guidance

Step 3: Scope Confirmation

After codebase-analyzer returns, confirm the design scope with the user before any design work. This is a recipe-local confirmation step.

Execute Skill: requirement-convergence before running the hearing protocol.

First run the requirement-convergence hearing protocol, using the codebase-analyzer findings as the facts it presents. In this flow, the orchestrator elicits and judges the fields and records the result as the skill's convergence object (outcome, requirements[] with layer labels, nonGoals[], plus a readiness label per field). Treat cost as already resolved because semantic repository investigation is assigned to codebase-analyzer and entering this recipe decided to design. Carry that object into Step 4 so technical-designer persists it to the Design Doc.

Then present, sourced from the codebase-analyzer JSON, using AskUserQuestion:

  • Target files/modules: analysisScope.filesAnalyzed and the modules they belong to
  • Affected layers: layers touched, derived from analysisScope.categoriesDetected and focusAreas
  • Unknowns/assumptions: limitations plus any assumptions codebase-analyzer recorded
  • Questions before design: open points that need a user answer before design proceeds

Ask the user to choose one:

  • Proceed to design with this scope — continue to Step 4 (Design Doc)
  • Correct the scope and re-run — return to Step 1 with the corrected scope; when the user names the corrected files or modules, use those directly as the Step 1 seed instead of re-deriving them by search
  • Hold additional hearing, then proceed — gather the missing answers, then continue to Step 4
  • Produce an ADR — when the confirmed scope involves architecture changes, new technology, or data flow changes, continue to Step 4 with technical-designer in ADR mode

After the user confirms the scope, count the confirmed target files and set the scale from the subagents-orchestration-guide Scale Determination table. This confirmed scale supersedes the Step 2 provisional value and determines the design document.

[STOP]: Wait for the user's choice before proceeding.

Step 4: Design Document Creation

Pass the full codebase-analyzer JSON to technical-designer (handoff contract HC-02). ADRs use alternative comparison; Design Docs use Design Convergence.

  • Invoke technical-designer using Agent tool
  • For Design Doc: subagent_type: "dev-workflows-fullstack:technical-designer", description: "Design Doc creation", prompt: "Create Design Doc based on the requirements. Requirements: [user requirements verbatim]. Codebase analysis: [codebase-analyzer JSON from Step 2]. Confirmed scope and user answers: [Step 3 confirmed scope and user answers]. Convergence result: [Step 3 convergence object]. Apply the code: prefix to codebase-analyzer fact_ids when filling the Fact Disposition Table."
  • For ADR: subagent_type: "dev-workflows-fullstack:technical-designer", description: "ADR creation", prompt: "Create ADR for [technical decision]. Requirements: [user requirements verbatim]. Codebase analysis: [codebase-analyzer JSON from Step 2]. Confirmed scope and user answers: [Step 3 confirmed scope and user answers]. Present at least two alternatives with trade-offs."
  • (Design Doc only) Invoke code-verifier to verify the Design Doc against existing code. Skip for ADR.
  • subagent_type: "dev-workflows-fullstack:code-verifier", description: "Design Doc verification", prompt: "doc_type: design-doc document_path: [Design Doc path] Verify Design Doc against existing code."
  • (Design Doc only) Invoke document-reviewer to verify consistency, completeness, and adopted design validity
  • Treat the preceding code-verifier result as code_verification evidence; the document-reviewer result controls correction routing.
  • subagent_type: "dev-workflows-fullstack:document-reviewer", description: "Design Doc review", prompt: "Review [Design Doc path] for consistency, completeness, and adopted design validity. doc_type: DesignDoc. review_context: creation. requirements_verbatim: [user requirements verbatim]. confirmed_decisions: [Step 3 confirmed scope and user answers]. codebase_analysis: [codebase-analyzer JSON from Step 2]. code_verification: [code-verifier output from this step]"
  • Apply the Review Resolution Gate. When apply findings change the Design Doc, invoke technical-designer in update mode, then re-run code-verifier and document-reviewer with prior_feedback.
  • (ADR only) Invoke document-reviewer to verify consistency and completeness
  • subagent_type: "dev-workflows-fullstack:document-reviewer", description: "ADR review", prompt: "Review [ADR path] for consistency and completeness. doc_type: ADR. codebase_analysis: [codebase-analyzer JSON from Step 2]"
  • Apply the Review Resolution Gate. When apply findings change the ADR, invoke technical-designer in update mode and re-run document-reviewer with prior_feedback.
  • (Design Doc only) Invoke design-sync to verify consistency across design documents. Skip for ADR-only.
  • subagent_type: "dev-workflows-fullstack:design-sync", description: "Design consistency check", prompt: "Check consistency across all Design Docs in docs/design/. Report conflicts and overlaps."
  • Apply the Review Resolution Gate to every reported conflict.
  • One or more apply findings → invoke the named technical designer for each affected Design Doc; re-run code-verifier and document-reviewer for each modified document with the latest verification and prior_feedback, then re-run design-sync
  • Every actionable conflict is decline → proceed to the approval stop
  • Any unresolved user_decision_required conflict → stop for user input

[STOP]: Present the design document, plus design-sync results for a Design Doc, and obtain user approval. For an approved ADR, invoke technical-designer in update mode to set its status to Accepted and verify the update before completion.

Completion Criteria

  • [ ] Built the Step 1 scope bootstrap seed (or obtained target files from the user when the search returned none)
  • [ ] Executed codebase-analyzer with a populated requirement_analysis
  • [ ] Ran the requirement-convergence hearing and carried its result into design
  • [ ] Confirmed the design scope with the user and set the scale from the confirmed target files
  • [ ] Created appropriate design document (ADR or Design Doc) with technical-designer
  • [ ] Executed code-verifier on Design Doc and passed results to document-reviewer (skip for ADR-only)
  • [ ] Executed document-reviewer and addressed feedback
  • [ ] Executed design-sync for consistency verification (skip for ADR-only)
  • [ ] Obtained user approval for the design document and verified an approved ADR has status Accepted

Output Example

Design phase completed.

  • Design document: docs/design/[document-name].md or docs/adr/[document-name].md
  • Approval status: User approved

How to use it

Copy the folder

Take shinpr/recipe-design from the repository into ~/.claude/skills for personal use, or into .claude/skills inside a project.

Check the name does not clash

The agent identifies a skill by the name field in its header. Two skills with the same name cannot sit side by side — one of them will be ignored.