SuperAPI AI Directory
Partner中文AI newsFavorites
Skillsmoneyprinter-v2

moneyprinter-v2

Agents & models

View repository

Updated 11-07 04:01

About this skill

MoneyPrinterV2 operator

Operate the named MPV2 application safely. This skill was audited against commit a734fa924adb9ffe5ab8472aa5526e469e42a8b5 (2026-09-15); the audited facts, hazards and pin live in references/upstream.md. Upstream describes the project as educational; the operator's job is to run it only on accounts and data the user owns.

When to use this skill

  • Fit-check, install, configure, preflight, run, schedule, or troubleshoot MoneyPrinterV2.
  • Work with its four menu areas: YouTube Shorts Automation, Twitter Bot, Affiliate Marketing, Outreach.
  • Explain config.json, the .mp/ cache, Post Bridge cross-posting, or why a scheduled job did not run.
  • Not for harry0703/MoneyPrinterTurbo (moneyprinter-turbo), the Maps scraper alone (google-maps-scraper), or general video editing.

Instructions

1. Freeze the surface

Pick one mode: orient, install, configure, run, schedule, or troubleshoot, and one feature area. Confirm the user owns the target accounts and has the right to publish the generated content. MPV2 and MoneyPrinterTurbo are different projects; confirm which one the user means.

2. Run the offline gate first

The helper reads host facts and, with --repo, a checkout. It never starts MPV2, Firefox, Ollama, Go or pip, opens no sockets, and writes nothing:

python3 .agent-skills/moneyprinter-v2/scripts/preflight.py --json
python3 .agent-skills/moneyprinter-v2/scripts/preflight.py --repo /path/to/MoneyPrinterV2 --config /path/to/config.json --json

An installed skill uses python3 "$SKILLS_ROOT/moneyprinter-v2/scripts/preflight.py". Output includes audit:* findings read from the checkout. A blocked repo.files result means the checkout differs from the audited commit; re-audit before trusting this skill's guidance.

3. Gate every side effect

Require separate explicit approval for each, in this order of escalation:

  1. pip install -r requirements.txt (third-party packages plus a wheel fetched from a GitHub release URL) into a virtualenv.
  2. bash scripts/setup_local.sh, which installs packages and rewrites config.json.
  3. Pulling Ollama models, and any Gemini image-API spend.
  4. Any run that drives a logged-in Firefox profile. MPV2 force-kills every running Firefox first; save browser work.
  5. Anything that publishes: YouTube upload, tweets, Post Bridge cross-posting.
  6. Scheduling, including OS cron or launchd entries that call src/cron.py.
  7. Outreach: downloading and building the Maps scraper with Go, scraping, crawling business websites for emails, and sending mail over SMTP.

Never copy a Firefox profile path's contents, SMTP password, API key, or .mp/ account cache into chat or logs.

4. Follow the matching reference branch

  • Install, Python, dependencies, setup_local.sh → Install and setup in references/upstream.md.
  • Config keys, env fallbacks, stale docs → Configuration.
  • YouTube, Twitter, affiliate flows → Feature areas.
  • Scheduled jobs that do not fire → Scheduling truth.
  • Maps scraping and email → Outreach chain and the responsible-use gate below.

5. Verify narrowly

Check the smallest effect first: python3 scripts/preflight_local.py inside the checkout (it makes network calls to Ollama and the image API, so it needs approval too), one dry generation without upload, then one upload. Report what ran and what did not; a green preflight is not proof of a working upload because the Twitter and YouTube flows depend on page selectors.

Responsible-use gate

Refuse or redirect requests that use MPV2 for unsolicited bulk email without a lawful basis and an opt-out, impersonation, fake engagement, evading platform detection, or posting under accounts the user does not control. For Outreach, ask what law and platform terms apply to the user's recipients and data before any scrape or send; collected business emails are personal data in many jurisdictions. AI-generated posts and videos may need disclosure under platform rules.

Examples

Fit check and gate

Ask which feature area and account the user means, run the preflight with --repo and --config, and report: Python 3.12 present, Ollama reachable class (loopback or remote), whether the image key is set, and the audit:* hazards. Do not install anything yet.

"My daily upload never ran"

Read Scheduling truth. The in-app CRON menu only registers jobs in a process that never runs them, so propose an OS scheduler calling src/cron.py <purpose> <account-id> <ollama-model> after approval, and note that scripts/upload_video.sh omits the model argument.

Best practices

  • Prefer the checkout's config.example.json and src/config.py over the prose docs; some docs list keys the code no longer reads.
  • Keep one account and one feature per session; test with headless: false so a person can watch.
  • Treat config.json as a secret file: it can hold SMTP credentials in plaintext.
  • Keep the helper offline and honest: a tool on PATH is not a working Ollama, Firefox login, or API key.

References