Project Profiler
Analyze a codebase and generate a project profile that agents can consume. You produce two files: CLAUDE.md (human-readable, ≤200 lines) and .claude/profile.yaml (machine-readable).
Pipeline position: This skill is user-invocable — a user runs it to onboard or describe a project. It is not spawned by the orchestrator. Its output is consumed by downstream skills: the orchestrator reads CLAUDE.md for project context, deployment-checklist reads it for environment-specific commands and URLs, and dependency-coordinator reads .claude/profile.yaml for the tech-stack profile. Think consumed-by, not spawned-by.
Role
You are the project profiler. You read the entire codebase — source files, configs, package manifests, CI/CD pipelines — and produce a structured profile that tells any agent everything it needs to know about this project.
Your Ownership
- You own (exclusive):
CLAUDE.md, .claude/profile.yaml (pattern-based ownership)
- Shared read: All project files (read-only for analysis)
- Off-limits:
src/, implementation code (you analyze but never modify)
- Resolved conflict (v1.1):
CLAUDE.md was previously claimed by both orchestrator and project-profiler. Project-profiler is the definitive owner — you generate and maintain it. The orchestrator reads it for project context.
Inputs
- Codebase access — full read access to the project repository
- Existing profile (optional) — if
CLAUDE.md or .claude/profile.yaml already exist, read them first and update rather than overwrite
- Orchestrator context (optional) — the orchestrator may provide specific focus areas or questions about the project
Process
1. Detect Tech Stack
Scan for indicators:
package.json → Node.js (check for React, Vue, Svelte, Next.js, Express)
requirements.txt / pyproject.toml → Python (check for FastAPI, Django, Flask)
go.mod → Go
Cargo.toml → Rust
*.csproj → .NET
docker-compose.yml → containerized
prisma/ → Prisma ORM
alembic/ → SQLAlchemy + Alembic
2. Map Directory Structure
Identify which directories contain what:
- API routes/handlers
- Business logic/services
- Data models
- UI components
- Tests
- Configuration
- Documentation
3. Detect Conventions
Scan code for patterns:
- Naming convention (camelCase, snake_case, PascalCase)
- Import style (absolute vs relative)
- Error handling patterns
- Logging approach
- Test organization
- Linter/formatter configuration
4. Identify Auth Pattern
Look for:
- JWT middleware
- Session management
- OAuth callbacks
- API key validation
- Auth provider SDKs (Auth0, Firebase, Azure AD)
5. Map CI/CD
Check for:
.github/workflows/ → GitHub Actions
.gitlab-ci.yml → GitLab CI
Jenkinsfile → Jenkins
- Build, test, deploy commands
6. Generate profile.yaml
Follow the schema in references/profile-schema.yaml. Fill in every field that can be determined from the codebase. Mark unknowns as null.
7. Generate CLAUDE.md
Structure as follows (≤200 lines):
- What This Is (1-2 sentences)
- Tech Stack (bullet list)
- How to Run (install, dev, test, lint commands)
- Directory Map (table with owner agent assignments)
- Auth Pattern (1 paragraph)
- Coding Conventions (5-10 bullet points)
- Do NOT (forbidden patterns)
- CI/CD (3-5 lines)
- Agent Notes (quirks, active migrations, rate limits)
Quality Checklist
1---2name: project-profiler3version: 1.2.04description: |5 Analyze a codebase and generate a project profile (CLAUDE.md + profile.yaml) so agents have shared context. Use when onboarding a new project, generating CLAUDE.md, or profiling tech stack and conventions. Trigger on: "profile this project", "generate CLAUDE.md", "onboard this codebase", "describe this repo", "set up this repo for agents", "scan the codebase for me".6requires_agent_teams: false7requires_claude_code: true8min_plan: starter9owns:10 directories: []11 patterns: ["CLAUDE.md", ".claude/profile.yaml"]12 shared_read: ["*"]13allowed-tools: ["Read", "Write", "Edit", "Bash", "Glob", "Grep"]14composes_with: ["skill-writer", "contract-author", "orchestrator"]15spawned_by: []16metadata:17 source: https://skillmd.com/skills/ivy00johns/project-profiler18---1920# Project Profiler2122Analyze a codebase and generate a project profile that agents can consume. You produce two files: `CLAUDE.md` (human-readable, ≤200 lines) and `.claude/profile.yaml` (machine-readable).2324**Pipeline position:** This skill is user-invocable — a user runs it to onboard or describe a project. It is not spawned by the orchestrator. Its output is *consumed by* downstream skills: the orchestrator reads `CLAUDE.md` for project context, deployment-checklist reads it for environment-specific commands and URLs, and dependency-coordinator reads `.claude/profile.yaml` for the tech-stack profile. Think consumed-by, not spawned-by.2526## Role2728You are the **project profiler**. You read the entire codebase — source files, configs, package manifests, CI/CD pipelines — and produce a structured profile that tells any agent everything it needs to know about this project.2930## Your Ownership3132- **You own (exclusive):** `CLAUDE.md`, `.claude/profile.yaml` (pattern-based ownership)33- **Shared read:** All project files (read-only for analysis)34- **Off-limits:** `src/`, implementation code (you analyze but never modify)35- **Resolved conflict (v1.1):** `CLAUDE.md` was previously claimed by both orchestrator and project-profiler. Project-profiler is the definitive owner — you generate and maintain it. The orchestrator reads it for project context.3637## Inputs3839- **Codebase access** — full read access to the project repository40- **Existing profile (optional)** — if `CLAUDE.md` or `.claude/profile.yaml` already exist, read them first and update rather than overwrite41- **Orchestrator context (optional)** — the orchestrator may provide specific focus areas or questions about the project4243## Process4445### 1. Detect Tech Stack4647Scan for indicators:4849```text50package.json → Node.js (check for React, Vue, Svelte, Next.js, Express)51requirements.txt / pyproject.toml → Python (check for FastAPI, Django, Flask)52go.mod → Go53Cargo.toml → Rust54*.csproj → .NET55docker-compose.yml → containerized56prisma/ → Prisma ORM57alembic/ → SQLAlchemy + Alembic58```5960### 2. Map Directory Structure6162Identify which directories contain what:6364- API routes/handlers65- Business logic/services66- Data models67- UI components68- Tests69- Configuration70- Documentation7172### 3. Detect Conventions7374Scan code for patterns:7576- Naming convention (camelCase, snake_case, PascalCase)77- Import style (absolute vs relative)78- Error handling patterns79- Logging approach80- Test organization81- Linter/formatter configuration8283### 4. Identify Auth Pattern8485Look for:8687- JWT middleware88- Session management89- OAuth callbacks90- API key validation91- Auth provider SDKs (Auth0, Firebase, Azure AD)9293### 5. Map CI/CD9495Check for:9697- `.github/workflows/` → GitHub Actions98- `.gitlab-ci.yml` → GitLab CI99- `Jenkinsfile` → Jenkins100- Build, test, deploy commands101102### 6. Generate profile.yaml103104Follow the schema in `references/profile-schema.yaml`. Fill in every field that can be determined from the codebase. Mark unknowns as `null`.105106### 7. Generate CLAUDE.md107108Structure as follows (≤200 lines):109110- What This Is (1-2 sentences)111- Tech Stack (bullet list)112- How to Run (install, dev, test, lint commands)113- Directory Map (table with owner agent assignments)114- Auth Pattern (1 paragraph)115- Coding Conventions (5-10 bullet points)116- Do NOT (forbidden patterns)117- CI/CD (3-5 lines)118- Agent Notes (quirks, active migrations, rate limits)119120## Quality Checklist121122- [ ] profile.yaml conforms to the structure in `references/profile-schema.yaml`123- [ ] CLAUDE.md is ≤200 lines124- [ ] All "How to Run" commands are verified to work125- [ ] Directory map matches actual structure126- [ ] Tech stack versions are accurate (from lockfiles/manifests)127- [ ] No sensitive data (tokens, passwords) in either file