Adapting Claude Pipeline

Adapting Claude Pipeline to a Codebase

Overview

Adapt the generic .claude/ pipeline folder to a specific project's tech stack, domain, and workflows. This means modifying, replacing, or removing every skill, agent, hook, script, prompt, and setting so they serve the target codebase — not the original Laravel/web template.

Core principle: Every file in .claude/ must earn its place. If it doesn't serve the target project, delete it. If it needs modification, modify it. If new capabilities are needed, create them.

When to Use

• Cloning the claude-pipeline repo into a new project
• Onboarding an existing codebase with the generic .claude/ folder
• Switching tech stacks (e.g., Laravel to Python, web to CLI, monolith to microservices)

The Process

dotCopy code
digraph adaptation {
    rankdir=TB;

    "1. Brainstorm with user" [shape=box style=filled fillcolor=lightyellow];
    "2. Research domain patterns" [shape=box];
    "3. Audit existing .claude/ inventory" [shape=box];
    "4. Write adaptation plan" [shape=box];
    "5. Execute plan with subagents" [shape=box style=filled fillcolor=lightblue];
    "6. Verify adapted pipeline" [shape=box style=filled fillcolor=lightgreen];

    "1. Brainstorm with user" -> "2. Research domain patterns";
    "2. Research domain patterns" -> "3. Audit existing .claude/ inventory";
    "3. Audit existing .claude/ inventory" -> "4. Write adaptation plan";
    "4. Write adaptation plan" -> "5. Execute plan with subagents";
    "5. Execute plan with subagents" -> "6. Verify adapted pipeline";
}

Phase 1: Brainstorm

REQUIRED: Use the brainstorming skill to explore the target project.

Focus questions on:

• Tech stack: Language, framework, database, infrastructure
• Project type: Web app, CLI tool, library, API, microservice, data pipeline, etc.
• Team workflows: CI/CD, code review, deployment, testing patterns
• Pain points: What does the team struggle with that agents could help?
• Existing tooling: Linters, formatters, test runners, build systems
• Scope: Full adaptation or partial (e.g., keep orchestration, replace agents)?

Key decisions to reach:

1. Which existing skills/agents/hooks are relevant (keep/modify)?
2. Which are irrelevant (delete)?
3. What new capabilities does the project need?
4. What domain-specific best practices should agents follow?

Phase 2: Research Domain Patterns

Use WebSearch to investigate the target project's domain. This feeds directly into agent definitions and skill content.

Search for:

Copy code
"[language/framework] best practices [current year]"
"[language/framework] anti-patterns"
"[language/framework] code review checklist"
"[language/framework] common mistakes"
"[language/framework] testing best practices"
"[language/framework] security considerations"
"[language/framework] project structure conventions"
"[tool/framework] CI/CD pipeline best practices"

Capture findings as notes — they will be incorporated into:

• Agent anti-patterns sections
• Skill content (technique skills)
• Hook logic (formatting, linting)
• Orchestration script stages

Phase 3: Audit Existing Inventory

Categorize every .claude/ file into one of four buckets:

Inventory Checklist

Skills (19 in template):

Agents (10 in template):

Hooks:

Settings (settings.json):

Scripts:

Prompts:

Phase 4: Write the Plan

REQUIRED: Use the writing-plans skill.

The plan should be organized into parallel workstreams where possible:

Copy code
Workstream A: Delete irrelevant files (quick, no dependencies)
Workstream B: Modify existing files (parallel per file)
Workstream C: Create new skills (use writing-skills skill)
Workstream D: Create new agents (use writing-agents skill)
Workstream E: Modify orchestration scripts (use cc-orchestration-writer agent)
Workstream F: Update hooks and settings

Each task in the plan must specify:

• File path to create/modify/delete
• What changes and why
• Which skill/agent to use for the task (writing-skills, writing-agents, cc-orchestration-writer, etc.)
• Dependencies on other tasks (if any)

Phase 5: Execute with Subagents

REQUIRED: Use the subagent-driven-development skill to execute the plan.

Route tasks to the correct agent/skill:

Parallel execution: Tasks in different workstreams with no shared files can run in parallel using dispatching-parallel-agents.

Phase 6: Verify

After all modifications:

1. File inventory check — Glob .claude/ and verify no orphaned or irrelevant files remain
2. Cross-reference check — Grep for references to deleted skills/agents/files and fix broken refs
3. Settings validation — Ensure settings.json hooks reference existing files and commands
4. Skill description audit — All skill descriptions match new project context
5. Agent coordination audit — Deferral relationships between agents are consistent
6. Dry run — Walk through a typical workflow (e.g., implement-issue) mentally to verify the pipeline makes sense

Common Adaptation Patterns

Web to CLI/Library

Delete: bulletproof-frontend, review-ui, ui-design-fundamentals, bulletproof-frontend-developer agent, all frontend prompts

Replace: laravel-backend-developer → language-specific developer agent

Modify: code-simplifier (change language), test validators (change framework)

Laravel to Python/Django

Replace: laravel-backend-developer → django-backend-developer, php-test-validator → pytest-validator, code-simplifier → python code simplifier, write-docblocks → python docstring writer

Modify: settings.json (Pint → Black/Ruff), post-pr-simplify.sh (PHP → Python)

Monolith to Microservices

Add: Service-specific agents, API contract validation skill, cross-service testing patterns

Modify: implement-issue workflow (multi-repo awareness), code-reviewer (service boundaries)

Adding New Domain

Add: Domain-specific agents (ML engineer, data pipeline developer, etc.), domain testing skills

Keep: All process skills, orchestration infrastructure

Red Flags

• Keeping irrelevant skills "just in case" — delete them. They add noise and confuse skill discovery.
• Generic agent personas — every agent should reference the actual project, its structure, and its conventions.
• Orphaned references — grep for deleted skill/agent names in remaining files.
• Skipping web research — domain patterns and anti-patterns are critical for agent quality.
• Not updating settings.json — stale hooks that reference missing tools will cause errors.

Key Principles

• Delete aggressively — fewer, focused files beat a bloated pipeline
• Research before writing — WebSearch for domain patterns before creating agents/skills
• Test the pipeline — walk through a real workflow after adaptation
• Use the meta-skills — writing-skills and writing-agents exist to ensure quality; use them for new content
• Preserve process skills — brainstorming, TDD, debugging, planning are universal; don't delete these