filetree

中文版

A Claude Code plugin that maintains FILETREE.md — a one-line description per file with content hashes for staleness detection. Lets the LLM grasp repo layout in a few hundred tokens before touching code.

Why

Every new Claude Code session relearns the codebase: ls, grep, open file, read, repeat. The discovery is expensive and not reusable across sessions.

Pain point	filetree's answer
LLM keeps rediscovering layout each session	Persisted as FILETREE.md, checked into git, shared with collaborators
Summary docs go stale silently	Per-entry content hash; mismatch reveals drift immediately
Rewriting every description on small refactors wastes tokens	UNCHANGED bias — LLM refreshes hash only when purpose is intact (~100x cheaper)
Tooling adds a sqlite / daemon / watcher	Single markdown file. Change detection delegated to git. Zero background process

Install

Via marketplace:

/plugin marketplace add nekocode/filetree-skill
/plugin install filetree

Local development / dog-fooding (no install needed):

cd /path/to/filetree-skill
claude --plugin-dir .

After editing commands/ or SKILL.md, run /reload-plugins inside the session for a hot reload.

Commands

Command	Purpose
/filetree:init	Generate FILETREE.md from scratch. Refuses to overwrite without confirmation
/filetree:update	Sync FILETREE.md with current repo state (added / changed / removed / renamed)
/filetree:lint	Read-only drift check. Exits non-zero on drift, CI-friendly. Does not call the LLM

All commands refuse to commit FILETREE.md. You review the diff and commit yourself.

Wire it into CLAUDE.md / AGENTS.md

/filetree:init handles this on first run — it scans the repo root for CLAUDE.md / AGENTS.md, skips files that already have a ## FILETREE.md section, and for the rest proposes a dedicated ## FILETREE.md section (rendered in the file's language). You confirm each edit before it lands.

Caveats:

• If neither CLAUDE.md nor AGENTS.md exists, the plugin won't create one — it's your call which (if any) to seed. Create the file you want, then re-run /filetree:init.
• Wiring runs at init time only. If you add CLAUDE.md / AGENTS.md later, re-run /filetree:init (it'll ask before overwriting FILETREE.md) or wire by hand.

To wire by hand, drop a section like this into your CLAUDE.md:

## FILETREE.md

`FILETREE.md` indexes every file with a one-line role summary, grouped by directory. Read it to find code by purpose before `ls`/`grep`/`find` — it turns "search the repo" into "look it up". Auto-maintained; don't hand-edit.

```
## (root)/

- `manage.py`: Django CLI entrypoint

## src/auth/

- `jwt.py`: JWT middleware; parses token, injects user_id
- `session.py`: Redis-backed server-side session store
```

The agent then treats FILETREE.md as a cheap index — one read replaces dozens of ls / grep / cat calls during orientation.

Project config (.filetree.json)

Optional. Drop a .filetree.json at the repo root and commit it to share with the team. Absent → defaults.

{
  "manifest_path": "docs/FILETREE.md",
  "exclude": ["migrations/", "**/*.gen.ts", "/build"],
  "include": ["*.svg"],
  "language": "zh",
  "commit_guard": true
}

Key	Effect	Default
manifest_path	Where the manifest is written (relative path inside the repo)	FILETREE.md
exclude	gitignore-style patterns to keep tracked files OUT of the manifest	[]
include	gitignore-style patterns to index files normally skipped (e.g. *.svg)	[]
language	Pin the summary language (e.g. "zh") instead of auto-detecting	null
commit_guard	Intercept a Claude-issued git commit and auto-update FILETREE.md	false

exclude / include accept full gitignore syntax (/build, **, !keep.gen.ts, trailing-slash dirs). Invalid config fails fast with a clear error.

Manifest format

# Project Filetree

_Auto-maintained by `/filetree:update`. Content hashes live in the sidecar `FILETREE.hash.json`; do not edit it by hand._

## (root)/

- `README.md`: Project entry doc

## src/auth/

- `middleware.py`: JWT validation middleware; parses bearer token and injects user_id into request context
- `jwt_utils.py`: Pure JWT signing / verification helpers, framework-agnostic

• Section heading ## dir/ = full directory path; root files under ## (root)/
• File line - `name`: summary — pure prose, no inline noise; an agent reads a file's location straight off its section heading
• Sections lexical, files within lexical → no spurious diffs
• Content hashes are stored out-of-band in FILETREE.hash.json ({path: hash}), keeping the manifest free of per-line hex noise. A pre-sidecar manifest with inline <!--hash:--> is auto-migrated on the next update.

Compatibility

Requirement	Version	Notes
git	any modern release	Required at runtime; non-git repos fail fast with a clear error
python3	≥ 3.9	Uses PEP 585 list[dict] builtin generics. Stdlib only — zero third-party deps. The plugin invokes python3 (the python command is absent on modern macOS / fresh Linux)
Claude Code	any	Plugin format. claude is shipped as a native binary; Node is not required

Development

# install pytest if you don't have it
python3 -m pip install pytest pytest-cov

# run tests
python3 -m pytest tests/ -q

# with coverage (target: 100% lines)
python3 -m pytest tests/ --cov=filetree --cov-report=term-missing

Tests load the script via importlib (see tests/conftest.py), so no package install is needed.

Lint your own FILETREE.md while iterating:

python3 skills/filetree/scripts/filetree.py lint

Exit code 1 = drift, 0 = clean. Wire it into pre-commit or CI as needed:

# .github/workflows/filetree.yml
- run: python3 skills/filetree/scripts/filetree.py lint

License

MIT. See .claude-plugin/plugin.json.