Skip to content

Guild-Agents/guild / .claude/skills/guild-specialize

guild-specialize

Verified

Enriches CLAUDE.md by exploring the project and specializes agents to the real stack

Guild Specialize

Explores the user's real project and enriches the entire Guild configuration with concrete information about the detected stack, architecture, and conventions.

This skill runs once after guild init. It transforms generic placeholders into real project information.

When to use

  • Immediately after running guild init
  • When a new stack is added to the project (new database, new framework)
  • When the project structure has changed significantly

Process

Step 1 — Read base context

Read the Guild configuration files:

  • CLAUDE.md — current instructions (contains [PENDING: guild-specialize] placeholders)
  • PROJECT.md — identity and stack declared during init

Step 2 — Explore the real project

Investigate the real project structure looking for:

Dependencies and versions:

  • package.json (Node.js/frontend)
  • pom.xml or build.gradle (Java)
  • requirements.txt or pyproject.toml (Python)
  • Gemfile (Ruby)
  • go.mod (Go)
  • Cargo.toml (Rust)

Architecture and structure:

  • Folders src/, app/, lib/, pkg/, internal/
  • Organization pattern: by layers, by features, by domain
  • Project entry points

Configuration and conventions:

  • tsconfig.json, eslint.config.*, .prettierrc
  • .env.example, .env.local (environment variables — do NOT read real .env)
  • Dockerfile, docker-compose.yml
  • CI/CD: .github/workflows/, .gitlab-ci.yml

Database and migrations:

  • Folder migrations/, db/, prisma/, drizzle/
  • Configured ORM or query builder
  • Existing schema

Existing documentation:

  • README.md — project overview
  • Internal documentation in docs/

Step 3 — Enrich CLAUDE.md

Invoke the Tech Lead agent using Task tool with model: "opus" (reasoning tier) to replace all [PENDING: guild-specialize] placeholders in CLAUDE.md with real information:

  • Stack with exact versions: extracted from dependency files
  • Folder structure explained: what each main folder does
  • Detected code conventions: linter, formatter, import style
  • Identified architecture patterns: MVC, hexagonal, modular, etc.
  • Known environment variables: listed from .env.example
  • Visible limitations and technical debt: outdated dependencies, TODOs found
  • Useful project commands: detected npm/make/cargo scripts

CLAUDE.md now uses zone markers (<!-- guild:auto-start:ID --> / <!-- guild:auto-end:ID -->) to delimit auto-generated sections. When enriching:

  • Replace content BETWEEN the markers, preserving the markers themselves
  • The following zones exist: structure, architecture, conventions, env-vars
  • Do NOT modify content outside of zone markers (user-owned sections)
  • If markers are missing (legacy project), replace [PENDING: guild-specialize] placeholders directly

Step 4 — Specialize agents

Invoke the Tech Lead agent using Task tool with model: "sonnet" (execution tier) to add project-specific context for each agent in .claude/agents/*.md:

  • advisor.md: real project domain, target users
  • tech-lead.md: specific stack, detected patterns, architecture decisions
  • developer.md: code conventions, main framework, file structure
  • code-reviewer.md: lint rules, project patterns, anti-patterns to watch
  • qa.md: testing framework, commands to run tests, current coverage
  • bugfix.md: debugging stack, logs, available tools

When specializing agents, append a zone at the bottom of each agent file:

<!-- guild:auto-start:agent-context -->
## Project-Specific Context
- Stack: [detected stack]
- Architecture: [detected patterns]
- Conventions: [detected conventions]
<!-- guild:auto-end:agent-context -->

This zone allows guild-re-specialize to update agent context later without touching the agent's core role definition.

Use the Task tool with model: "sonnet" to invoke each agent by reading their .claude/agents/[name].md if you need a specialized perspective to enrich their configuration.

The model parameter is resolved from the step's model-tier: reasoning→"opus", execution→"sonnet". System/gate steps run inline (no Task tool).

Step 5 — Confirm

Present a summary of what was detected:

Guild v1 specialized for [project-name]

Detected stack:
- [list of technologies with versions]

Architecture:
- [identified pattern]
- [main structure]

Updated agents:
- [list of agents with their applied specialization]

Use /resume or /build-feature to continue work.

Step 6 — Commit enrichment immediately

CRITICAL: After enriching CLAUDE.md and agent files, commit the changes immediately as their own atomic commit. Do NOT leave them as unstaged changes — they are vulnerable to git stash and other operations.

git add CLAUDE.md .claude/agents/*.md
git commit -m "chore: enrich CLAUDE.md and agents via guild-specialize"

This ensures enrichment survives any subsequent git operations (stash, checkout, rebase).

Example Session

User: /guild-specialize

Guild Specialize analyzing project...

Tech Lead (opus) — Enriching CLAUDE.md...
Stack detected:
- Node.js 20.11.0, TypeScript 5.3.3
- React 18.2.0, Next.js 14.1.0
- PostgreSQL via Prisma 5.9.0

Architecture:
- Next.js App Router (src/app/)
- API routes in src/app/api/
- Shared components in src/components/

Tech Lead (sonnet) — Specializing agents...
Agents updated:
- developer.md: Specialized for Next.js + TypeScript
- qa.md: Configured for Vitest + Playwright

Use /resume or /build-feature to continue work.

Important Notes

  • NEVER read real .env files — only .env.example or .env.local
  • If you cannot detect something with certainty, ask the user instead of assuming
  • Prioritize accuracy over completeness — it is better to say "not detected" than to fabricate
  • Agents should be specialized to the real stack, not generic
  • NEVER use git stash in automated pipelines — use wip: commits instead
  • CLAUDE.md changes must always be committed separately from feature code
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 465kUpdated Claude CodeCodexVerifiedInstall
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 465kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 319kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 319kUpdated Claude CodeCodexVerifiedInstall