skills/leveldb-ops/SKILL.md
Read and inspect LevelDB stores - especially Chromium/Electron app state (Local Storage, IndexedDB, Session Storage). Triggers on: leveldb, .ldb files, IndexedDB, Local Storage, Chromium storage, Electron app state, claude.ai cache, browser forensics, decode app state, claude desktop state.
npx skillsauth add 0xDarkMatter/claude-mods leveldb-opsInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
Read and decode LevelDB stores — primarily the Chromium/Electron storage layers (Local Storage, IndexedDB, Session Storage) used by every Electron app on disk: Claude Desktop, VS Code, Discord, Slack, Obsidian.
Embedded key-value store by Google. Sorted KV map, no SQL, no server. Format: a folder of .ldb (sorted runs), .log (write-ahead), MANIFEST-*, CURRENT, LOCK. Both keys and values are arbitrary bytes.
Chromium layers richer formats on top:
LevelDB uses an exclusive LOCK file. A running app holds it. Trying to open a live store fails OR silently returns stale snapshots.
Always copy before reading:
# Copy the entire leveldb dir to a temp location
cp -r "$APPDATA/Claude/Local Storage/leveldb" /tmp/probe/local-storage-db
cp -r "$APPDATA/Claude/IndexedDB/https_claude.ai_0.indexeddb.leveldb" /tmp/probe/indexeddb
# Remove the copied LOCK file so the reader can open it
rm -f /tmp/probe/local-storage-db/LOCK /tmp/probe/indexeddb/LOCK
The cp -r will warn Device or resource busy for the LOCK file itself — that's fine, the data files copy successfully.
Never write to the live store while the app is running. It will corrupt the LSM and crash the app. Quit the app first if you need to mutate.
plyvel and similar require native compilation and lack Windows wheels. Use ccl_chromium_reader — pure Python, written for browser forensics.
uv venv .venv --python 3.13
source .venv/Scripts/activate # or .venv/bin/activate on Unix
uv pip install "git+https://github.com/cclgroupltd/ccl_chrome_indexeddb.git"
Not on PyPI — install direct from GitHub. Pulls in ccl-simplesnappy and brotli as transitive deps.
Storage keys are origin URLs (https://claude.ai). Records are append-only — duplicate script_key entries mean older versions; the last record wins.
import pathlib
from ccl_chromium_reader import ccl_chromium_localstorage
ls = ccl_chromium_localstorage.LocalStoreDb(pathlib.Path("./local-storage-db"))
# List all origins
for origin in sorted(set(ls.iter_storage_keys())):
print(origin)
# Dump one origin, latest-value-wins
latest = {}
for rec in ls.iter_records_for_storage_key("https://claude.ai"):
latest[rec.script_key] = rec.value
for k, v in latest.items():
print(f"{k}: {repr(v)[:200]}")
See scripts/dump_localstorage.py for the full reusable script.
IndexedDB is more complex — wrapped object stores with v8-serialized values. ccl_chromium_reader parses it cleanly:
from ccl_chromium_reader import ccl_chromium_indexeddb
db = ccl_chromium_indexeddb.WrappedIndexDB(pathlib.Path("./indexeddb"))
for db_id in db.database_ids:
wdb = db[db_id.dbid_no]
print(f"DB: {wdb.name}")
for store_name in wdb.object_store_names:
store = wdb[store_name]
for rec in store.iterate_records():
print(f" {rec.key!r} -> {repr(rec.value)[:200]}")
See scripts/dump_indexeddb.py.
| OS | Path |
|----|------|
| Windows | %APPDATA%\<App>\Local Storage\leveldb\ |
| Windows | %APPDATA%\<App>\IndexedDB\https_<host>_0.indexeddb.leveldb\ |
| macOS | ~/Library/Application Support/<App>/Local Storage/leveldb/ |
| Linux | ~/.config/<App>/Local Storage/leveldb/ |
For raw browsers, <App> is Google/Chrome/User Data/Default, BraveSoftware/Brave-Browser/User Data/Default, etc.
Writing requires either:
level package, or rebuild the dir manually) — or--remote-debugging-port=<n> flag, attach, and call localStorage.setItem(key, value). Survives the app's normal write path so it doesn't corrupt the LSM.For Claude Desktop specifically, see references/claude-desktop-state.md for the discovered key map.
| You want to | Do |
|-------------|-----|
| Just see what's there | Copy + ccl_chromium_reader |
| Find a specific value | strings + grep first; reader if structure matters |
| Mutate while app runs | Don't. Use DevTools remote debugging. |
| Mutate while app is closed | Quit, then Node level package or write back via re-opened leveldb |
| Cross-account recovery | Read-only forensics; can't impersonate server-bound entries |
strings for structured analysis → misses keys, conflates duplicates, can't distinguish originstesting
Audit any repo against the agentic-quality doctrine — score entry docs, structure, and enforcement gates, then map each finding to its fix. Triggers on: repo doctor, repo audit, agentic quality, is this repo agent-friendly, doc drift, stale AGENTS.md, monorepo structure, nested CLAUDE.md.
data-ai
Router for parallel or recurring agent work across six skills. Covers: parallel agents, fan out work, delegate to workers, run overnight, scheduled loop, land branches, mixed-model fleet, orchestrate workers, background agents at scale. Triggers on: which skill for parallel work, fan out agents, spawn workers, run this overnight, schedule a loop, land my branches, heterogeneous fleet, delegate to cheaper model, autonomous loop.
tools
Heterogeneous cross-provider fleet - GLM (z.ai), Codex (OpenAI), Grok (xAI), Anthropic Sonnet/Opus/Haiku - from one session, porting the native Workflow tool's patterns (adversarial verify, judge panels, journal resume) to OS-process workers. Triggers: fleetflow, heterogeneous/mixed-model fleet, codex worker, grok worker, cross-provider fan-out, cross-model verify.
development
Application/game-scale three.js: ES modules, GLTF pipeline (DRACO/KTX2/meshopt), AnimationMixer, physics (rapier/cannon-es), react-three-fiber, and performance at scale (InstancedMesh, LOD, draw calls). Triggers on: three.js, GLTFLoader, r3f, game loop, WebGL memory leak, boids.