agentleFS
LongJohnSilver1504LongJohnSilver1504/claude-skills/skills/plan-implementation/SKILL.md

plan-implementation

Skill1 starsChanged 57 days agoMITView on GitHub

Bridge design artifacts (PRD, UX spec with embedded test matrix) into a structured implementation plan. Classifies deliverables as domain features or shared infrastructure, extracts specs, orders by dependencies, and detects shared patterns. Use when a PRD/UX spec is ready to become a build plan — after prd-to-ux, when planning feature scaffolding, or when user says "implementation plan", "what to build first", "plan the features".

8 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

11 rules. 0 are common, 8 are distinctive, and 3 are very specific to this project.

  • 0

    Common

    Most of the 8 edited versions we read keep these.

  • 8

    Distinctive

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    When 5+ deliverables exist, include a mermaid dependency graph.

    line 74
  2. Distinctive:

    Before finalizing, scan for deliverables that will independently use the same logic:

    line 78
  3. Distinctive:

    Do 2+ hooks independently fetch the same data?

    line 80
  4. Distinctive:

    Do 2+ hooks independently transform data through the same pipeline?

    line 81
  5. Very specific:

    Do 2+ hooks independently extract the same value from router/context/store?

    line 82
  6. Distinctive:

    If found: extract a shared hook/utility as a new deliverable, insert it before the consumers in the implementation order, update consumers …

    line 84
  7. Distinctive:

    Before writing the plan, review every deliverable spec for vague instructions.

    line 90
  8. Distinctive:

    Every spec item must include:

    line 104
  9. Distinctive:

    If you cannot make a spec item concrete, it needs to be broken down further or clarified with the PRD/UX spec.

    line 109
  10. Very specific:

    Read the feature's existing artifacts (.claude/pipeline/{feature}/ and the feature folder: DESIGN.md, PRD.md, UX-spec.md, the …

    line 132
  11. Very specific:

    Use the AskUserQuestion tool for the transition (never numbered text options), with a one-line summary of what was completed.

    line 136

“Common” is measured: kept by at least 6 in 10 of the 8 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Plan Implementation

Bridge between design artifacts and code scaffolding. Read PRD and UX spec (which includes the test matrix in Pass 9). Produce a structured plan where each deliverable maps to either:

  • create-feature — domain features with entities, API endpoints, CRUD (lives in features/)
  • create-infrastructure — shared infrastructure: providers, hooks, layouts, i18n, config (lives in shared/)

Required Inputs

At minimum: the PRD. More artifacts = more precise extraction.

  1. PRD — the product requirements document
  2. Clarified PRD — if prd-clarifier was used (optional)
  3. UX Specification — from prd-to-ux (includes test matrix in Pass 9)

Process

Step 1: Classify Candidates

Scan design artifacts and extract every distinct deliverable. Classify each using these signals:

SignalDomain FeatureInfrastructure
Has a primary entity with fieldsYesNo
Consumes external API endpointsYesRarely
Has CRUD operationsYesNo
Lives infeatures/{name}/shared/, public/, root
Examplespayment, reservationauth-provider, app-layout, i18n

Where to find candidates:

From PRD: Section 4 (core use case steps → features), Section 5 (capabilities → features or infra), Section 7 (data sources → entities vs config).

From UX Spec: Pass 2 (concept groups → features), Pass 5 (elements with state tables → features), layout components → infrastructure.

Output a classification table with: name, source reference, type (Domain Feature / Infrastructure), primary concept.

If ALL candidates are infrastructure (no entities, no CRUD), state explicitly that create-feature does not apply and all deliverables use create-infrastructure.

Step 2: Extract Specs

For each candidate, produce a spec block.

Domain Feature spec — include: feature name (singular, lowercase), entity fields with types, API endpoints with HTTP methods, which CRUD operations are needed, UX components mapped to build prompts, whether a Zustand store is needed and why, domain-specific errors.

Contract verification (required per API endpoint): for every endpoint in a domain feature spec, include a verified-against field stating what the request/response shape was checked against — one of: real API, MSW mock, backend doc, or UNVERIFIED. Zod schemas drifting from real API responses is a recurring bug class; UNVERIFIED endpoints must be flagged prominently in the plan so implementers and reviewers know the schema is a guess, not a contract.

Infrastructure spec — include: name, type (one of the Deliverable Type Catalog entries in the create-infrastructure skill: Provider, Shared hook, Layout component/hook/barrel, i18n namespace, Shared utility, Config/Foundation, Test page), target file path, dependencies on other deliverables, acceptance criteria, props/API surface, exports.

Step 3: Identify Shared Infrastructure

List everything that spans multiple features or must exist before scaffolding. Always include:

  • Routes to add (centralized links)
  • shadcn components to install
  • Dependencies to install
  • API client changes needed

For infrastructure-only projects, this section IS the main content.

Step 4: Order by Dependencies

Order all deliverables so dependencies come first.

Infrastructure-first rule: shared infrastructure always before domain features that depend on it.

Include: order number, deliverable name, type, depends on, complexity estimate, what can be parallelized.

When 5+ deliverables exist, include a mermaid dependency graph.

Step 4.5: Detect Shared Patterns

Before finalizing, scan for deliverables that will independently use the same logic:

  • Do 2+ hooks independently fetch the same data?
  • Do 2+ hooks independently transform data through the same pipeline?
  • Do 2+ hooks independently extract the same value from router/context/store?

If found: extract a shared hook/utility as a new deliverable, insert it before the consumers in the implementation order, update consumers to import it instead of duplicating.

This prevents DRY violations that are expensive to fix later.

Step 4.7: Enforce Concrete Specs (No Placeholders)

Before writing the plan, review every deliverable spec for vague instructions. Every spec must have concrete, specific instructions — never placeholders.

Placeholder (reject these):

  • "Add appropriate error handling"
  • "Implement the remaining methods"
  • "Add validation as needed"
  • "Style the component appropriately"

Concrete (require these):

  • "Add handleApiError with status 404 mapped to RESERVATION_NOT_FOUND"
  • "Create useReservationList hook that calls reservationApi.list(), returns { data, isPending, error }"
  • "Add Zod schema: z.object({ phone: z.string().min(10) })"
  • "Use Card with bg-card, text-card-foreground, rounded-lg border"

Every spec item must include:

  • Exact file paths to create or modify
  • What it produces (a function, a component, a hook, a type)
  • How to verify (test command, or "file compiles")

If you cannot make a spec item concrete, it needs to be broken down further or clarified with the PRD/UX spec.

Step 5: Write the Plan

Save as {prd-basename}-implementation-plan.md in the same directory as the source PRD, following references/PLAN-FORMAT.md (source paths, summary table, classification, shared infrastructure, ordered spec blocks).

Done when: the plan file exists and every item in the Validation Checklist below is checked.

Validation Checklist

  • Every PRD functional requirement maps to at least one deliverable
  • Every deliverable is classified as Domain Feature or Infrastructure
  • Domain features have: entity fields with types, endpoints, operations
  • Every API endpoint has a contract-verification field (real API / MSW mock / backend doc / UNVERIFIED)
  • Infrastructure deliverables have: type, location, dependencies, acceptance criteria
  • Implementation order respects dependencies
  • Shared infrastructure listed first
  • If infrastructure-only, plan states create-feature does not apply
  • Routes use centralized links (not hardcoded)
  • Shared patterns detected and extracted

Resume After Context Cleanup

Read the feature's existing artifacts (.claude/pipeline/{feature}/ and the feature folder: DESIGN.md, PRD.md, UX-spec.md, the implementation plan, PROGRESS.md — whichever exist), then this skill's own input, and continue from where they leave off — never restart from scratch.

Next Step

Use the AskUserQuestion tool for the transition (never numbered text options), with a one-line summary of what was completed. Options:

  • execute-tasks (primary, default) — invoke /execute-tasks with the path to the implementation plan. Recommend starting a fresh session first (/clear) — the plan IS the context.
  • Manual scaffolding — only when the user explicitly wants one deliverable by hand: /create-feature (domain feature) or /create-infrastructure (infrastructure deliverable) with the spec
  • Something else — do something different

Related Skills

SkillRelationship
prd-to-uxUpstream — produces UX spec with embedded test matrix consumed here
execute-tasksDownstream (primary) — consumes the plan and executes it with subagents
create-featureDownstream (manual alternative) — scaffolds one domain feature by hand
create-infrastructureDownstream (manual alternative) — scaffolds one infrastructure deliverable by hand
react-clean-architectureReference — informs layer decisions during extraction

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.