Skill: map-codebase

Purpose: Scan the codebase, identify all major features/flows, and produce a code-map.md that traces each feature from entry point → routing → logic → DB. Makes any codebase immediately navigable without grepping.

Trigger Phrases

• "map the codebase"
• "build code map" / "create code map"
• "map my project"
• "analyze codebase structure"
• "create a roadmap of the code"
• "where does everything live"

Steps

Step 1 — Detect Tech Stack

Glob and read top-level files to identify the stack and project layout:

Glob("**/{package.json,pom.xml,build.gradle,requirements.txt,setup.py,*.sln,web.xml}", limit=20)

Then read the top-level directory listing to understand the folder structure. Identify:

• Language: Java, JavaScript/Node, Python, .NET, Go, Ruby, etc.
• Frontend type: HTML + vanilla JS, React, Vue, JSP templates, etc.
• Routing pattern: Servlet if/else chain, Express routes, Django urlpatterns, Flask decorators, Rails routes
• DB access pattern: Raw SQL, JPA/Hibernate, Sequelize, SQLAlchemy, ActiveRecord

Step 2 — Find Entry Points

Scan for where requests come in:

Also look for:

• API spec files (openapi.yaml, swagger.json) — pre-mapped routes
• URL pattern files (.claude/rules/url-patterns.md if it exists)

Step 3 — Find Major Logic Files

Identify the heavy files that contain most of the business logic:

Bash: find files > 300 lines
Grep: look for files imported/required by many others

Files > 500 lines in services/, utils/, jobs/, lib/ are almost always key logic files. Read their method list (grep for public , def , function , const .*=.*=>) to understand their surface area.

Step 4 — Find DB Access

Grep: "FROM ", "INSERT INTO", "UPDATE ", "CREATE TABLE" → extract table names
Grep: "models.Model", "@Entity", "@Table", "Schema(" → extract model names
Glob: "migrations/", "*.sql", "*_install.sql", "*schema*" → schema files

Build a list of all DB tables/models referenced in the logic files.

Step 5 — Identify Feature Groups

Group entry points + logic + DB by feature domain. Look for naming patterns:

• user, auth, login, session → Auth/User flow
• order, cart, checkout, payment → Commerce flow
• admin, dashboard, manage → Admin flow
• course, lesson, session, email, scheduler → Course/content flow
• api, webhook, export → Integration flow

Aim for 5–15 feature groups depending on codebase size. Don't try to map every file — focus on the flows a developer actually needs to navigate.

Step 6 — Trace Each Flow

For each identified feature group, read 2–3 key files to trace the full chain. Document:

Entry (URL / route / page)
  → Frontend (HTML element / JS function)
  → Endpoint / Route handler
  → Service / Logic method [file:approxLine]
  → DB tables read/written

Use Grep for function names and Read with offset to confirm logic without loading whole files.

Step 7 — Build the Line Reference

For the 1–2 largest logic files (the ones developers will need to navigate most), build a method → line number table:

Grep: "public |def |function |  async " in the file → capture method names + line numbers

This is the highest-value part of the map for large files (500+ lines).

Step 8 — Write code-map.md

Write to .claude/rules/code-map.md (create .claude/rules/ if it doesn't exist).

Format:

# [Project Name] — Code Map
> Generated by map-codebase. Re-run after major structural changes.
> `[Abbreviation]` = `path/to/main/logic/file`

---

## [Feature Name]
- **Entry** `URL or route`
- **Frontend** `file.js` → `functionName()` → `callbackName()`
- **Endpoint** `routeName` → `HandlerClass.method()`
- **Logic** `ServiceClass.method()` [L####]
- **DB** tables/models read and written

## [Feature Name 2]
...

---

## Key File Paths
| Layer | Path |
|---|---|
| [description] | `path/to/file` |

---

## [MainLogicFile] — Line Reference
| Method | Line | Purpose |
|---|---|---|
| `methodName` | #### | one-line description |

Step 9 — Wire Into CLAUDE.md

Ask the user: "Add @rules/code-map.md to CLAUDE.md so it auto-loads each session?"

If yes: Edit .claude/CLAUDE.md to add @rules/code-map.md in the @rules/ block.

If no: Tell them they can read it manually anytime or re-run map-codebase to refresh it.

Notes

• Line numbers drift — they're navigation hints, not guarantees. The function names are what matters.
• Focus on the dominant paths — a flow that's touched in 10 places is more important than one that's touched in 1.
• Largest logic files first — if one file has 5000+ lines, the line reference table for that file is worth more than tracing all the smaller flows.
• Re-run when: new major feature added, large refactor done, new developer joining.
• Lite version: if the codebase is very large (50+ files), do Step 7 only for the top 2 files and summarize the rest as a file-path table.