The Steinberger Sweep — steipete's GitHub, inventoried and mined

2026-07-07 · Fable integration agent · 138 repos judged (60 steipete + 78 openclaw org) · winners installed with evidence

The one-paragraph verdict: Peter Steinberger's public work splits into three strata: (1) the crawlkit crawler suite (telecrawl / discrawl / gitcrawl / wacrawl) — local-first SQLite mirrors of your own communication archives, which is exactly the ingestion layer this system's brain wants; (2) agent-reach CLIs we already run daily (gog v0.9.0, wacli 0.12.0 — both were his all along); (3) the OpenClaw platform itself, which the user runs and which is out of scope by standing rule. We adopted the four crawlers that work on this machine, imported the user's full Telegram history read-only (45,968 messages / 200 chats) and WhatsApp Desktop history read-only (107,589 messages / 377 chats) into local searchable mirrors, and documented the human-gated auth steps for the rest.

1 · What landed (with receipts)

Version evidence (fresh, this session):

telecrawl : 0.3.1
discrawl  : 0.11.4
gitcrawl  : 0.7.0
wacli     : 0.12.0
wacrawl   : 0.3.1
gog       : v0.9.0 (99d9575 2026-01-22T04:15:12Z)
brew tap  : steipete/tap, openclaw/tap
brew list : discrawl gitcrawl telecrawl wacli wacrawl

2 · Deep dives — the named eight

telecrawl — Telegram → SQLite (ADOPTED, import verified)

Reads Telegram Desktop's on-disk tdata (and macOS Postbox) read-only and mirrors chats/messages/media metadata into ~/.telecrawl/telecrawl.db. No API login needed for the local-import path. Import ran on this machine, 2026-07-07 03:14:20Z → 03:20:23Z:

source_path: /Users/thescoho/Library/Application Support/Telegram Desktop/tdata
db_path:  /Users/thescoho/.telecrawl/telecrawl.db
chats: 200          messages: 45968
media_messages: 11360   topics: 268
oldest_message: 2022-02-21T06:19:19Z
newest_message: 2026-07-07T03:06:37Z

Four years of Telegram history is now greppable offline (telecrawl search <query>). CORRECTION (2026-07-07, verified against openclaw/telecrawl README + telecrawl --help): v0.3.1 has NO live API sync and never uses my.telegram.org api_id/api_hash — the original claim here was wrong. Freshness path: re-run telecrawl import after Telegram Desktop has synced; --fetch-media reuses local session auth data without any login flow.

discrawl — Discord → SQLite (ADOPTED, sync human-gated)

Mirrors Discord servers/channels/messages into a local SQLite cache; can also wiretap the local client cache. Installed at 0.11.4. Live bot sync requires a Discord bot token the human must create (§5). No token was typed or stored by this agent.

gitcrawl — GitHub issues/PRs → SQLite (ADOPTED)

Local-first crawler for GitHub issues and pull requests with clustering/triage tooling — rides the existing gh auth, so it works today. Smoke test:

$ gitcrawl metadata --json | head
{ "schema_version": "crawlkit.control.v1", "id": "gitcrawl",
  "display_name": "Git Crawl",
  "description": "Local-first GitHub issue and pull request crawler." … }

wacrawl — WhatsApp Desktop → SQLite (ADOPTED, import verified)

Crawls the WhatsApp Desktop app's local data at ~/Library/Group Containers/group.net.whatsapp.WhatsApp.shared. The earlier skip was wrong/stale: this machine now has /Applications/WhatsApp.app (net.whatsapp.WhatsApp, version 26.23.72) plus the expected local databases. wacrawl doctor verified ChatStorage.sqlite, ContactsV2.sqlite, and Message/Media, then wacrawl status imported a read-only snapshot:

source_path: /Users/thescoho/Library/Group Containers/group.net.whatsapp.WhatsApp.shared
db_path: /Users/thescoho/.wacrawl/wacrawl.db
chats: 377          messages: 107589
contacts: 531      groups: 55
participants: 2678 media_messages: 7100
oldest_message: 2021-03-22T02:02:53Z
newest_message: 2026-07-07T03:59:05Z
last_import: 2026-07-07T04:00:13Z

wacrawl is read-only and does not use the QR/websocket auth path. It complements wacli: wacrawl is the offline archive mirror; wacli is live send/read via WhatsApp Web pairing.

wacli / gogcli (ALREADY-HAVE — hands off)

Both were already in the toolbelt before this sweep; the sweep's contribution is provenance — they're steipete tools, same author as the crawler suite. wacli was upgraded from 0.2.0 to 0.12.0 after wacli auth failed with Client outdated (405) connect failure; that failure was an outdated WhatsApp Web client, not missing WhatsApp Desktop data. gog is in daily Gmail use; per run contract its config/OAuth were not touched.

crawlkit (NOTED — the foundation)

The shared Go library beneath every crawler: SQLite storage conventions (crawlkit.control.v1), snapshotting, backup, and a shared TUI. Nothing to install by itself; its crawlctl scheduler is the future piece if we ever want cron-driven re-crawls.

crabbox (SKIP-FOR-NOW)

Remote test/exec control plane — a coordinator + runner fleet for farming out commands. Architecturally interesting for Mission Control's cross-machine story, but we run no runner fleet; adopting it today would be infra cosplay. Revisit if remote runners become real.

3 · What this unlocks (FUTURE notes — no config was changed)

• Brain ingestion: telecrawl's DB is a ready corpus for ~/brain/inbox/ / gbrain ingestion — a telecrawl search … --json → markdown digest pipeline is a one-evening build. Same shape for discrawl once a bot token exists.
• Agent-reach complement: agent-reach reaches out to 13 platforms; the crawlers mirror your own archives locally. Together: outward search + inward memory, both offline-capable.
• OpenClaw channel potential: discrawl/telecrawl DBs could back OpenClaw channel context (history-aware replies) — noted as future only; ~/.openclaw/* was not touched, per contract.
• gitcrawl triage: point it at the repos this system owns to get offline issue/PR triage with clustering — a natural feed for the Dreaming Ledger's "what needs attention" surface.

4 · Full inventory — steipete org (60 repos judged)

Verdict key: ADOPT installed with evidence · ALREADY-HAVE in toolbelt pre-sweep · QUEUE adopt on trigger · NOTED read/mine, don't install · SKIP reasoned pass.

5 · Full inventory — openclaw org (78 repos judged)

6 · Human-gated auth checklist (exact commands, nothing attempted by the agent)

Per contract, no credentials were typed, echoed, or logged; no QR/login flow was attempted.

7 · Verification ledger

• Inventory: gh api 'users/steipete/repos?per_page=100' --paginate (+ same for orgs/openclaw) → /tmp/steipete-repos.jsonl; verdicts in /tmp/verdicts-{steipete,openclaw}.tsv, joined into the tables above.
• npm/pipx squatting check: not applicable to the winners — crawler binaries were installed from trusted Homebrew formulae under steipete/tap and openclaw/tap, with release artifacts hosted at github.com/openclaw/*.
• Installs: version strings pasted in §1 were captured this session, not transcribed from READMEs.
• Telegram import: read-only against live tdata; telecrawl rate-limit backoffs (9–12s) observed and survived; counts in §2 are from telecrawl status post-import.
• Honest nulls: crabbox is not installed — reasons in §2; crawlctl scheduling untested; discrawl end-to-end sync unverified until a human supplies a bot token.

Usage doctrine: ~/.claude/skills/steipete-crawlers/SKILL.md (verified commands only).