SKILL.md 9.6 KB


name: init-project description: Initialize a complete project from scratch. Plugin check → interview → analyze → design → validate → scaffold skeleton → plan v1 features → validate plan → implement (TDD, subagents) → analyze → review → finish. Same implementation rigor as ship-feature. argument-hint: disable-model-invocation: true

allowed-tools: Read, Write, Edit, Bash, Grep, Glob

ORCHESTRATOR: INIT PROJECT

AGENTS AND SKILLS LOADED

Custom agents (this config):

  • .claude/agents/plugin-advisor.md ← plugin configuration check
  • .claude/agents/interviewer.md ← project questionnaire
  • .claude/agents/analyzer.md ← risk/constraint analysis
  • .claude/agents/scaffolder.md ← project skeleton (NO features, skeleton only)

Superpowers skills (design + implementation):

  • superpowers:brainstorming ← architecture design
  • superpowers:writing-plans ← v1 feature decomposition
  • superpowers:subagent-driven-development ← isolated TDD implementation
  • superpowers:requesting-code-review ← final review
  • superpowers:finishing-a-development-branch ← cleanup + verification

INITIAL REQUEST

$ARGUMENTS


WORKFLOW


STEP 0 — PLUGIN CHECK

Load and follow: .claude/agents/plugin-advisor.md

Feed it the initial request above.

If ACTION REQUIRED: YES:

================================================================
⚠️  PLUGIN CHECK — ACTION REQUIRED
================================================================
[paste full RECOMMENDATIONS block]
----------------------------------------------------------------
A) Enable recommended plugins then re-run /init-project
B) Type "force" to proceed without them
================================================================

STOP. Wait for user response.

  • Re-run → restart from STEP 0
  • "force" → note missing plugins, continue to STEP 1

If ACTION REQUIRED: NO: Print one line and continue: ✅ Plugin check passed — [active plugins]


STEP 1 — INTERVIEWER

Load and follow: .claude/agents/interviewer.md

Identify what is already provided in the initial request. Ask only what is genuinely missing. Single structured block of questions.

MANDATORY STOP — do not continue until user has answered.

Produce the PROJECT BRIEF. This is the single source of truth for all subsequent steps.


STEP 2 — ANALYZER

Load and follow: .claude/agents/analyzer.md

Analyze the PROJECT BRIEF:

  • Existing repo or code to integrate
  • Stack constraints and compatibility issues
  • Infrastructure constraints
  • Risks that could affect the design
  • Open decisions from the PROJECT BRIEF

Produce an ANALYSIS REPORT.


STEP 3 — ARCHITECTURE DESIGN

Invoke skill: superpowers:brainstorming

Feed it the PROJECT BRIEF + ANALYSIS REPORT.

Produce a complete DESIGN covering:

  • Finalized tech stack with exact versions
  • Complete folder structure (full tree)
  • Module responsibilities and data flow
  • Key interfaces and data models (signatures only — no implementation)
  • Config and tooling setup
  • Test strategy
  • Any open decisions resolved with justification
  • Prerequisites list (what must be installed on dev machine)

STEP 4 — VALIDATION GATE #1 — ARCHITECTURE

MANDATORY STOP — present to the user and wait for approval.

================================================================
INIT PROJECT — ARCHITECTURE VALIDATION
================================================================

PROJECT SUMMARY
---------------
<3–5 line recap of what will be built>

STACK
-----
<finalized stack with versions>

PREREQUISITES TO INSTALL
-------------------------
<list of tools / runtimes / services with versions>

FOLDER STRUCTURE
----------------
<full tree from the DESIGN>

V1 FEATURES TO IMPLEMENT
-------------------------
<numbered list from PROJECT BRIEF — these will be implemented
 in the pipeline AFTER the skeleton is scaffolded>

CONVENTIONS
-----------
<naming, doc style, test strategy>

EXCEPTIONS TO GLOBAL RULES
---------------------------
<list or "none">

================================================================
Approve this architecture? (yes / request changes)
================================================================

IF changes → return to STEP 3. IF approved → proceed.


STEP 5 — SCAFFOLD SKELETON

Load and follow: .claude/agents/scaffolder.md

Pass to the scaffolder:

  • Full PROJECT BRIEF
  • Approved DESIGN
  • ~/.claude/templates/project-CLAUDE.md
  • ~/.claude/CLAUDE.md

The scaffolder creates:

  1. CLAUDE.md — filled from global template, no placeholders
  2. .claude/settings.json — adapted to this stack
  3. .claudeignore — extended for this project
  4. Complete folder structure
  5. Config files (package.json, Cargo.toml, etc.)
  6. Empty entry points and module files (structure only, no business logic)
  7. .gitignore, .env.example

The scaffolder does NOT create the README and does NOT implement any features. README is handled by readme-updater (STEP 5b). Features are handled by the implementation pipeline (STEPs 6–9).

The scaffolder must verify: git init + build passes on empty project.


STEP 5b — CREATE README

Load and follow: .claude/agents/readme-updater.md

Context: README.md does not exist yet → CREATE mode activates automatically.

The readme-updater reads CLAUDE.md, the folder structure, and manifests to generate the full README (About, Prerequisites with OS-specific install commands, Installation, Running, Project structure, Configuration, Contributing).

No stop required — prints confirmation and continues immediately.


STEP 6 — PLAN V1 FEATURES

Invoke skill: superpowers:writing-plans

Using the PROJECT BRIEF v1 features list and the scaffolded skeleton as context:

  • Break each v1 feature into granular tasks (2–5 min each)
  • Each task must reference exact file paths from the scaffolded structure
  • Each task must include: what to implement, expected behavior, verification steps
  • Apply TDD: tests are written before implementation code

STEP 7 — VALIDATION GATE #2 — IMPLEMENTATION PLAN

MANDATORY STOP — present the plan to the user.

================================================================
INIT PROJECT — IMPLEMENTATION PLAN VALIDATION
================================================================

SKELETON STATUS : ✅ build passes
V1 FEATURES     : <N> features → <M> tasks

<numbered task list with file paths>

================================================================
Approve this plan and start implementation? (yes / request changes)
================================================================

IF changes → return to STEP 6. IF approved → proceed.


STEP 8 — IMPLEMENT V1 FEATURES

Invoke skill: superpowers:subagent-driven-development

Execute each task with isolated subagents. Mandatory TDD: superpowers:test-driven-development applies. Two-stage review per task: spec compliance → code quality.

Each subagent works on a clean context with:

  • The task description
  • Relevant file paths
  • PROJECT BRIEF context
  • CLAUDE.md conventions

STEP 9 — ANALYZE

Load and follow: .claude/agents/analyzer.md

Run the ANALYZER on the completed implementation:

  • Verify no regressions
  • Verify no plan deviations
  • Verify no stale scaffold code left (empty files not yet populated)
  • Verify conventions are respected

STEP 10 — CODE REVIEW

Invoke skill: superpowers:requesting-code-review

Full review scope:

  • Code quality vs CLAUDE.md conventions and global rules
  • Security issues
  • Missing or incomplete v1 features
  • README accuracy
  • Generated CLAUDE.md completeness (no placeholder remaining)
  • Test coverage

Fix all CRITICAL issues before proceeding.


STEP 11 — FINISH

Invoke skill: superpowers:finishing-a-development-branch

Verify:

  • All tests pass
  • Build is clean
  • No leftover scaffold placeholders
  • Initial commit prepared

STEP 12 — SYNC README

Load and follow: .claude/agents/readme-updater.md

Context: call with argument "sync".

SYNC mode — no stop required. The readme-updater:

  • Detects any drift between README and the implemented code
  • Updates commands, env vars, folder structure if changed during implementation
  • Adds a ## Recent changes entry summarizing v1 features
  • Prints one-line confirmation

RULES

  • Never skip STEP 0 — plugin check is mandatory.
  • Never skip STEP 1 — no assumptions about missing info.
  • Never implement without explicit user approval at STEP 4.
  • Never implement without explicit user approval at STEP 7.
  • The scaffolder only creates the skeleton — zero business logic.
  • All feature implementation goes through the subagent pipeline (STEP 8).
  • Apply CLAUDE.md norms throughout.
  • A broken skeleton or a broken build is not acceptable output.

FINAL OUTPUT

================================================================
PROJECT INITIALIZED: <project name>
================================================================

LOCATION    : <project root>
STACK       : <finalized stack>
BUILD       : ✅ / ❌ <e>
TESTS       : ✅ <N> passing / ❌ <detail>

V1 FEATURES
-----------
✅ <feature>
✅ <feature>
⚠️ <feature> — partial: <reason>

REMAINING ISSUES
----------------
<IMPORTANT and MINOR issues, or "none">

QUICK START
-----------
<exact commands to run the project right now>

CLAUDE.md  : ✅ complete
README.md  : ✅ created (STEP 5b) + synced (STEP 12)
SETTINGS   : ✅ .claude/settings.json + .claudeignore generated
================================================================