Claude Code Bootstrapper

Turn any project into an AI-augmented engineering harness
Works with any stack • any language Edition of April 2026 Claude Code • cc-bootstrapper
Everything Claude needs to know about your project lives right here.
committed to git
gitignored / local
auto-invoked
isolated context
router / entry point
§

Project Root

🗃 your-project/
├─ CLAUDE.md commit team instructions — architecture, rules, commands
├─ CLAUDE.local.md local personal overrides, gitignored
└─ ⚙ .claude/ the control center
├─ settings.json commit permissions + config
├─ settings.local.json local personal permissions
├─ agents/ isolated subagent personas
├─ skills/ auto auto-invoked workflows
├─ pipelines/ auto task routers
├─ scripts/hooks/ auto event-driven hooks
├─ memory/ facts, patterns, decisions, sessions
├─ database/ schema dump
├─ input/ tasks, plans, images
└─ output/ generated artifacts
◊ • ◊ • ◊
†

Agents

Subagent personas with isolated context. Each is a .md prompt in agents/.

  • analyst.md reads code & schema, creates specs
  • {lang}-architect.md plans modules & structure
  • {lang}-developer.md writes production code
  • {lang}-test-developer.md generates unit tests
  • {lang}-reviewer.md reviews architecture, logic & security
  • storage-architect.md DB, cache, queues design
  • devops.md infra, CI/CD, Docker
  • qa-engineer.md test plans, Postman, E2E
Pattern: one file = one persona. The agent gets its own isolated context window — it sees only what you give it in the prompt plus the tools it’s allowed.
‡

Skills

Auto-invoked workflows. Each is a directory with SKILL.md inside skills/.

  • pipeline/ task classifier & router
  • p/ alias for /pipeline
  • code-style/ enforces conventions
  • architecture/ module structure & patterns
  • testing/ test patterns & frameworks
  • storage/ DB, cache, queue patterns
  • memory/ 3-level memory system
  • gitlab/ github/ playwright/ conditional MCP skills
Pattern: skills are triggered automatically by context match or via /skill-name. Each has a SKILL.md with description + instructions.
◆

Pipelines

/pipeline {name} or /p {name}

Orchestration files in pipelines/. Each .md describes a multi-step workflow that chains skills and agents.

  • new-code.md write new feature
  • fix-code.md diagnose & fix bugs
  • review.md per-lang parallel review
  • tests.md generate tests
  • hotfix.md urgent hotfix
  • brainstorm.md multi-agent brainstorm
  • full-feature.md end-to-end cycle
  • api-docs.md API documentation
  • qa-docs.md QA checklists
Click any pipeline to read its full specification.
◊ • ◊ • ◊
→

Request Flow

Every code-related request is classified and routed through the pipeline system.

I
Request
user prompt or /command
II
Pipeline
classifies intent, selects workflow
III
Skill
loads context and orchestration rules
IV
Agent(s)
isolated personas execute in parallel
V
Result
code, review, docs, or artifacts

Hooks

Shell scripts in scripts/hooks/ that fire on Claude Code lifecycle events.

Event Fires when…
SessionStart Session begins — inject context, load state
PreToolUse Before a tool call — validate, gate, transform
PostToolUse After a tool call — log, track, audit
Stop Session ends — summarize, persist state
Notification Background event — alerts, side-effects
※

Memory & Context

Persistent knowledge that survives across sessions.

  • memory/ persistent knowledge across sessions
  • memory/facts.md stack, key paths, active decisions, known issues
  • memory/patterns.md code patterns & conventions
  • memory/issues.md recurring issues tracker
  • memory/decisions/ architecture decision records
  • memory/sessions/ session summaries
  • database/ schema snapshots, references
  • input/ task specs, plans, images
  • output/ generated artifacts
Pattern: memory/ survives between sessions. Hooks and pipelines update it via CAPTURE phases with REPLACE/MERGE semantics.
>_

See It In Action

claude — cc-bootstrapper