homelab
Skill4 starsChanged 5 months agoMITView on GitHub
Set up or manage the homelab workers on a residential-IP box that the autocomp loop reaches through Supabase queues — worker 1 (trawl-worker, a CAPTCHA/bot-wall fetch+solve tier) and worker 2 (agentic browser for interactive tasks: sign-ups, form fills, "log in and do X"). Worker 2's DEFAULT is the loop driving the captured Camoufox boseclaw session itself via `browser/camoufox.mjs act` (no API key, no daemon); the browser-use `agent-worker/` daemon is an optional always-on fallback. Use when the user says "/homelab", "set up the homelab worker", "get TRAWL/the agent running at home", "do the directory signups", or wants to stand up / check / restart either worker.
- Reads credentials
2 other projects use a version of this.
Adopted by
Most recently taken up by
- jameswyse/skillstheir own version, first seen Nov 15, 2026
- heartshare/claude-skillstheir own version, first seen Dec 13, 2025
Other versions2 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 2
- heartshare/claude-skills★ 1same name
- jameswyse/skills★ 1same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository13 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- autocomp-tickSkill
- ceoSkill
- fetch-protectedSkill
- initiate-ceoSkill
- initiate-ctoSkill
- initiate-gtmSkill
- initiate-qaSkill
- launch-directoriesSkill
- next-moveSkill
- plan-big-execute-smallSkill
- plan-small-execute-bigSkill
- solve-captchaSkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 1 is distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 16
If invoked on the VPS, say so and stop — a datacenter IP defeats the purpose (see tools/trawl.md "Pull-worker mode" and private/memory / …
- Very specific: line 21
Locate the bundle: If trawl-worker/setup.sh isn't in the current tree, clone the repo first: git clone …
- Very specific: line 34
Prove it end-to-end: Run the live test (./setup.sh --test, or answer yes at the prompt): it enqueues a real Cloudflare-walled page and …
- Distinctive: line 86
Only stand this up when you want unattended pickup that survives with no Claude session open (a true 24/7 headless drainer).
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
homelab — stand up the residential-IP solver worker
The autocomp loop runs on a datacenter VPS whose IP is flagged by CAPTCHA vendors and
which is too small to host a real browser. So the solver (TRAWL = Camoufox Firefox) runs
on the owner's homelab and the loop reaches it through a pull queue (Supabase
solve_jobs). This skill stands that worker up and manages it. Everything lives in
trawl-worker/ — the muscle is trawl-worker/setup.sh; this skill drives it and verifies
the result end-to-end.
Run this ON the homelab box (the residential-IP machine with Docker), not on the VPS.
If invoked on the VPS, say so and stop — a datacenter IP defeats the purpose (see
tools/trawl.md "Pull-worker mode" and private/memory / the trawl-pull-worker note).
Default action — set it up
- Locate the bundle. If
trawl-worker/setup.shisn't in the current tree, clone the repo first:git clone git@github.com:Eastkap/autocomp-live.git && cd autocomp-live. - Prereqs. Confirm Docker + the Compose plugin are installed and the daemon is
reachable (
docker info). If missing, point the user at the install docs and stop — installing Docker is theirs to do. - Gather the two secrets. The worker needs
SUPABASE_URLandSUPABASE_SECRET_KEY(both come from the autocomp.env). If the user hasn't provided them, ask for them — the secret key is hidden on input. Never echo the secret back or write it anywhere buttrawl-worker/.env. - Run the installer:
cd trawl-worker && ./setup.sh(interactive) — or, if you already hold the two values in the environment,SUPABASE_URL=… SUPABASE_SECRET_KEY=… ./setup.sh --yes --test. It writes.env(chmod 600),docker compose up -d --build, waits for TRAWL health, and confirms the worker is polling. - Prove it end-to-end. Run the live test (
./setup.sh --test, or answer yes at the prompt): it enqueues a real Cloudflare-walled page and waits for the worker to solve it. A green "LIVE SOLVE OK" means the whole path works: VPS → queue → homelab → TRAWL → back. Report the result honestly — if the solve fails, showdocker compose logs worker/logs trawl, don't declare success. - Tell the loop. Once the live solve passes, the owner should reply 'up' to the loop (or you, if you're the loop, flip the homelab kanban card to done) so the CAPTCHA-blocked directory signups resume automatically.
Management verbs
- status / is it working: from any box with the autocomp
.env,tools/solve.sh workersshows which worker claimed recent jobs; on the homelab,docker compose ps. - logs:
./setup.sh logs(follows the worker) ordocker compose logs -f trawl. - restart / update: re-run
./setup.sh(idempotent — pulls newer images, recreates). - reconfigure creds:
./setup.sh --reconfigure. - stop:
./setup.sh down. - tune throughput: raise
BROWSER_POOL_SIZEin.env(more concurrent Camoufox browsers — needs more RAM), then re-run./setup.sh.
Worker 2 — agentic browser (drive a real logged-in session)
The second lane: act in a real logged-in browser — sign-ups, form fills, "log in and do
X" the fetch tier can't. Identity ladder: boseclaw by default; only ever run
identity=personal on a box you trust, and enqueue it only for safe, reversible tasks
(spend/send/destructive stay gated upstream).
Default path — the loop drives the captured Camoufox session (no daemon, no API key)
This is the default and should be tried first. The loop is an agent (Claude Code), so it
doesn't need a second API-keyed agent to think for it — it just needs hands in a logged-in
browser. Those hands are browser/camoufox.mjs act, driving the boseclaw session captured by
login-capture.mjs (.secrets/boseclaw-state.json). No ANTHROPIC_API_KEY, no separate
Chrome profile, no fresh login — Claude authors the steps, watches the screenshots, iterates.
# one-time (already done for boseclaw): capture the session by hand in a real Camoufox window
node browser/login-capture.mjs boseclaw https://accounts.google.com/ # → .secrets/boseclaw-state.json
# then, for any interactive task — write a steps file and run it in the logged-in session:
node browser/camoufox.mjs act steps.json --state .secrets/boseclaw-state.json [--save-state .secrets/boseclaw-state.json]Steps are a JSON array (goto/waitFor/fill/type/click/select/press/wait/shot/text/url — see the
header of camoufox.mjs). It stops on the first failing step and screenshots so Claude can see
what happened and adjust. --save-state persists a new directory-account login back into the
session. CAPTCHA sites often pass on Camoufox's real fingerprint alone; when one truly blocks,
lean on worker 1 (the TRAWL solver) or hand the single blocked step to a human card. Consuming
the queue is just: read a queued agent_jobs row → drive it with act → PATCH it done/failed
with a result (verify live before claiming done, CLAUDE.md hard rules).
Optional fallback — the browser-use daemon (agent-worker/, always-on)
Only stand this up when you want unattended pickup that survives with no Claude session open
(a true 24/7 headless drainer). It's a separate Python daemon that calls the Anthropic API
itself, so it costs an ANTHROPIC_API_KEY (your spend) and wants its own pre-logged-in Chrome
profile (note: that's Chrome user_data_dir, not the Camoufox storage-state above — different
engine, so it needs its own one-time google-chrome --user-data-dir=…/profiles/boseclaw login).
Setup: cd agent-worker && cp .env.example .env (fill the 3 secrets + IDENTITIES), do the Chrome
login, docker compose up -d, then tools/agent-task.sh do "…" <url> boseclaw 900 to prove it.
worker.py pins a browser-use version — if its Agent API shifted, the first run's logs show it;
bump requirements.txt and re-test.
Guardrails
- Residential IP is the point — don't set this up on a cloud/datacenter box.
SUPABASE_SECRET_KEYis server-side; it stays intrawl-worker/.envon the homelab and is never sent to a browser, a prompt echo, a commit, or the kanban.- The worker is outbound-only — never instruct the user to open a port or expose 8191 to the
internet; it's published to
127.0.0.1for local debugging only. - Honest reporting: if the health check or live solve doesn't pass, say so with the logs — never report the worker as "up" unverified (CLAUDE.md hard rules + principle 13).
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

