● Live referencePublishing-only automationcontabo-149 · TLS auto
Social Media Multi-Task Manager
Control plane for browser-first social publishing: tool playbooks below, plus three live apps —
Agent Hub (code + prompts), Run Console (execute jobs),
OMNI (media daemons + cookie heal). Own-content only. Conservative pacing.
What: Library of 6 hybrid agent apps. Each page holds code, shell commands, and agent prompts for one job (publisher, stealth, growth, content, outreach, ops).
Hybrids: each agent wires 5–6 tools from this reference (e.g. AutoSocial + Cloak + FFmpeg + Aaron + CLI) into one runnable package — not a single-tool wrapper.
How to use: open an agent → copy the Python/shell runner → paste the agent prompt into your AI coding agent (Claude/Cursor/Grok) so it executes the code without manual steps. Or run the script on Contabo when deps are installed.
Best for: “I need a ready playbook + code for one role.”
What: Interactive dashboard. Pick platform + account, verify cookies server-side, auto-login if stale, then run a hybrid job and see PASS | FAIL | INCONCLUSIVE.
vs Agent Hub: Console executes live against Contabo vaults. Agent Hub documents & ships code/prompts for agents to run elsewhere.
How to use: open Console → select TikTok/X/IG/FB → choose action → Run. Passwords never leave the server; only status/logs return to the browser.
What: Media-ops control plane for launcher, auto-responder, keep-alive, overgrowth, auto-login, master_fix. Cookie verify + heal for multi-account stacks.
vs Run Console: Omni targets the OMNI media agent suite (ports/daemons). Console is the general hybrid-agent runner for publish/grow jobs.
How to use: open Omni → check API health → verify cookies → heal via auto-login → launch the media agent you need (launcher :8080, responder :8081, etc.).
Best for: “Keep sessions alive / open headful browsers / growth daemons.”
Agents hub vs hybrids (clear difference)
Agent Hub = code + prompts catalog
Six role pages under /agents/{name}/
Each includes runnable Python, shell, and a ready agent prompt
Hybrids = the 5–6 tools bundled inside that agent
Does not by itself click “publish” until something runs the code
How an AI agent runs them without you
Give the agent the page URL or paste the prompt block
Agent reads *_agent.py + commands, installs deps if needed
Agent executes steps and reports three-state results
Console/Omni APIs can be called by the same agent for live runs
Zero manual intervention pattern:
AI coding agent loads the agent prompt → runs hybrid code → uses Console API when live accounts are needed → logs PASS/FAIL/INCONCLUSIVE.
You only set secrets once on the server (cookies, credentials, token).
PublishPlaywright
Katzca/AutoSocial
Local dashboard for TikTok / IG / YouTube — persistent sessions, queues, schedule, FFmpeg uniquify.
Use: login once → queue own media → schedule. Hybrid input for Publisher agent.
A hybrid = one agent that combines 5–6 tools into a single runnable workflow. Open the agent page for code, commands, and the prompt your AI agent should follow.
Use when: multi-account day plan, IP routing, incident L1–L3 logging.
AI agent direct use (no manual clicks):
1) Point agent at /agents/<role>/ · 2) agent copies *_agent.py + prompt ·
3) agent runs commands and reports three-state · 4) for live accounts call /console/ API or open /omni/ heal path.
OMNI media agents — launcher · responder · keep-alive · overgrowth · auto-login · master_fix
(session/runtime plane — different from Agent Hub docs)
1) Draft captions (Aaron/Corey skills) · 2) Uniquify video (yt-dlp + FFmpeg / Apex) · 3) Queue AutoSocial · 4) Schedule windows · 5) Shadowban canary next day
Task board B — Multi-account
1) One profile dir per account · 2) Cookie inject before nav · 3) Residential proxy match geo · 4) Poisson delays · 5) Quarantine on checkpoint
Task board C — Growth day
1) Warm feed scroll · 2) Bounded engage · 3) No ban-loop retries · 4) L1 reduce / L2 re-export cookies / L3 48h halt
Task board D — Agent orchestration
1) simplified-cli JSON jobs · 2) marketing skills for copy · 3) Playwright only for own posts · 4) Log every exit code
Hard rule: Browser automation is for your own content publishing with conservative pacing.
No credential stuffing, no mass account creation, no captcha bypass automation (passive_only: true).
Daily multi-task matrix
Slot
Task
Tool
Notes
09:00
Session health
persistent profiles / cookies
refresh_silently ~45m on engage
10:00
Draft + score
Aaron/Corey skills
Virality ≥ 60 before publish if using growth gates
11:30
Queue publish
AutoSocial / auto-instagram
Own media only; human delays
14:00
Light engage
Playwright + stealth
PhasedGrowth caps if account < 21 days
18:00
Analytics + notes
simplified-cli / CRM
Three-state: PASS | FAIL | INCONCLUSIVE
Safety & Ethics
Use cautiously. Align with platform ToS and local law. This hub is a technical reference, not a green light for abuse.
Own content only
Publishing only for browser automation patterns — your media, your accounts.
Conservative pacing — Poisson burst-rest, Fitts delays, never 429 tight-loops.
One account → one VPS IP forever for live multi-account Mode C.
Quarantine on shadowban | checkpoint | session_expired | account_locked.
No captcha bypass automation — passive_only ethics.
CloakBrowser license: no illegal use, no credential stuffing, no account-creation abuse.
INCONCLUSIVE ≠ PASS. Never claim “done” without measurable verification. Do not weaken checks or hardcode success.
Katzca/AutoSocial — Playwright Social Dashboard
Local dashboard for TikTok / Instagram / YouTube with persistent Playwright sessions, queues, scheduling, yt-dlp + FFmpeg video uniquification.
First-run login per account — sessions saved to .profiles/.
Queue videos.
Schedule or instant post.
Extend with OpenCV / face-blurring logic when privacy requires.
Pro-tips
Keep one profile directory per account — never share cookies across brands.
Uniquify before multi-platform post: light crop, metadata strip, audio gain nudge via FFmpeg.
Prefer schedule windows that match audience timezone (pair with CloakBrowser geoip).
After post: monitor first 60 minutes engagement; freeze queue if rate limits appear.
Suggested commands
bash · clone & run (adapt to repo README)
git clone https://github.com/Katzca/AutoSocial.git
cd AutoSocial
# install deps per README (node/python)
# start dashboard, then login once per account
# sessions → .profiles/
# example FFmpeg uniquify (own content)
ffmpeg -i input.mp4 -vf "scale=1080:1920:force_original_aspect_ratio=increase,crop=1080:1920,eq=brightness=0.01:contrast=1.02" \
-c:v libx264 -preset fast -crf 20 -c:a aac -ar 48000 -shortest out_unique.mp4
Agent prompt
Using AutoSocial patterns: prepare a queue of N own-content videos for TikTok+IG.
- Ensure .profiles/{account} exists and is healthy
- Uniquify each video with FFmpeg (H.264, ar 48000)
- Schedule with ≥45m spacing
- Abort on checkpoint/shadowban signals
- Report PASS/FAIL/INCONCLUSIVE per account
xtea/auto-instagram — IG Publisher
Focused Instagram publisher (Feed / Carousel / Reels) using Playwright + imported cookies + Patchright (stealth Chromium). Queue support via cron.
Prepare a content folder (images/videos + captions).
Import cookies for the target account.
Run CLI commands for Feed / Carousel / Reels.
Add residential proxy + human-like delays.
Optional: cron for queue drain.
Clone
bash
git clone https://github.com/xtea/auto-instagram.git
cd auto-instagram
# import cookies → run CLI (see repo commands)
# add residential proxy + human delays
Cookie hygiene: encrypt at rest when possible; inject cookies before navigate; save after success; sanitize SameSite Titlecase; chmod 600 on secrets.
Agent prompt
Publish one Reel from ./content/reel-01 using auto-instagram patterns.
- Load cookies for account X
- Residential proxy + geo match
- Human delays between UI steps (Bezier clicks if available)
- Own content only; log final URL or FAIL reason
- Exit non-zero on challenge/checkpoint
Related Patterns (GitHub)
Other Playwright social patterns to study — not the default stack. Useful ideas for queues, multi-platform CLI, AI helpers.
cedonulfi/automie
Playwright + Gemini AI + Streamlit. Good for AI-assisted drafting + browser actions in one UI.
from cloakbrowser import launch
# Headed mode + residential proxy for maximum stealth
browser = launch(headless=False, proxy="http://your-residential-proxy:port")
page = browser.new_page()
page.goto("https://heavily-protected-site.com") # better odds vs DataDome etc.
browser.close()
This runs a real headed browser rendered on a virtual display — no physical monitor needed. Combine with geoip + humanize + residential IP.
Persistent context (cookies warm-up)
python
from cloakbrowser import launch_persistent_context
# First run: warm up with --disable-http2 (fresh sessions sometimes challenged over HTTP/2)
ctx = launch_persistent_context("./profile", args=["--disable-http2"])
page = ctx.new_page()
page.goto("https://example.com") # warms up cookies
ctx.close()
# Future runs — no --disable-http2 needed
ctx = launch_persistent_context("./profile")
page = ctx.new_page()
page.goto("https://example.com") # passes with saved cookies
javascript
import { launchPersistentContext } from 'cloakbrowser';
// First run: warm up with --disable-http2
let ctx = await launchPersistentContext({ userDataDir: './profile', args: ['--disable-http2'] });
let page = await ctx.newPage();
await page.goto('https://example.com');
await ctx.close();
// Future runs
ctx = await launchPersistentContext({ userDataDir: './profile' });
Avoid incognito penalty
python
from cloakbrowser import launch_persistent_context
ctx = launch_persistent_context("./my-profile", headless=False)
# If still flagged, raise storage quota (see CloakBrowser storage quota docs)
reCAPTCHA v3 — don’t use wait_for_timeout
python
# Bad — sends CDP commands, reCAPTCHA detects this
page.wait_for_timeout(3000)
# Good — invisible to the browser
import time
time.sleep(3)
# Prefer type() over fill()
page.type("#email", "user@example.com", delay=50)
javascript
// Bad — CDP
await page.waitForTimeout(3000);
// Good
await new Promise(r => setTimeout(r, 3000));
Use Playwright, not Puppeteer (less CDP noise for reCAPTCHA).
Residential proxies — datacenter IPs fail on reputation alone.
Spend 15+ seconds on page before triggering reCAPTCHA.
Space requests: 30+ seconds between grecaptcha.execute pages.
Fixed fingerprint seed for consistent device identity.
Minimize page.evaluate() before reCAPTCHA fires.
Troubleshooting (CloakBrowser / Playwright)
Most blocks come from missing residential proxy, geoip, or headed mode — not pure fingerprint magic.
Still blocked? (DataDome / Turnstile)
Some sites detect headless even with C++ patches. Use Xvfb + headed mode + residential proxy (see above).
Detected by FingerprintJS?
Detection
Cause
Fix
nodriver / bad bot
Stale binary/wrapper, missing FPJS patches, poor proxy reputation
Upgrade Pro binary (150.0.7871.114.3+), residential + geoip=True, config below
Browser tampering
Noise injection detected by ML
--fingerprint-noise=false
Browser tampering (fonts)
Font metrics ≠ spoofed Windows platform
--fingerprint-windows-font-metrics (Chromium 148+; Windows fonts installed)
On minimal Linux, missing font packages break canvas emoji hashes. Install the Font Setup on Linux packages from CloakBrowser docs after proxy/geoip/headed are correct.
Fresh session challenges over HTTP/2
Warm cookies once with --disable-http2, then reuse persistent profile. For ephemeral: launch(args=["--disable-http2"]) forces HTTP/1.1 — only when required.
Using simplified-cli:
1) Generate an image for campaign C with ai-image:generate --wait
2) Create a post draft with posts:create
3) Return raw JSON + exit code
4) If exit != 0 → FAIL with stderr
🥇 Aaron Marketing Skills
Best overall marketing OS for agents: 120 skills across 7 disciplines with deterministic quality gates.
Works with any Agent Skills client; cross-references; full marketing stack; agency-tested.
Cons
Not standalone; skills are instructions not scripts; needs product-marketing.md; can overwhelm for one narrow task.
bash
npx skills add coreyhaines31/marketingskills
# Fill .agents/product-marketing.md
# Then ask NL questions or use /cro /emails /seo-audit
Example agent asks
Optimize this landing page for conversions
Write a 5-email welcome sequence
Run /seo-audit on https://example.com
Draft IG captions for a product launch using product-marketing.md
eracle/OpenOutreach — B2B Email Agent
If priority is B2B sales outreach (not social). Autonomous lead discovery + AI email sequences from your own mailbox.
Copy-paste prompts for multi-task social management. Combine with skills above.
1 · Multi-account publish day
You are the social multi-task manager for manager.addict.best.
Accounts: [list]. Platforms: IG Reels + TikTok.
Constraints: own content only; residential proxies; one profile per account; ≥45m spacing;
quarantine on checkpoint/shadowban; three-state report.
Plan → uniquify → queue → publish → first_60_minutes checklist.
2 · Caption system (Aaron/Corey bind)
Using product-marketing context + social skill:
Write 5 platform-native captions (IG, X, LinkedIn, TikTok, YT short).
Emit SHIP/FIX/BLOCK for each with reasons. No banned claims. Include CTA + 3–8 tags max.
3 · Stealth recovery
Account hit soft block after Playwright post.
Diagnose using CloakBrowser troubleshooting matrix (proxy, geoip, headless, fonts, HTTP/2 warm).
Propose minimal fix; do not retry tight-loop; max 3 heal attempts with evidence.
4 · Growth vs risk
Account age 12 days. Propose PhasedGrowth caps for engage today.
Skip parasitic_risk tags. Include warm_up steps and stop conditions (301/308/310 map).
5 · Orchestrator shell
Build a bash/python orchestrator that:
- calls simplified-cli for assets
- writes queue JSON for AutoSocial/auto-instagram
- launches cloakbrowser persistent contexts
- logs PASS|FAIL|INCONCLUSIVE per step
Never hardcode PASS.
Command Cheatsheet
High-signal commands collected from the full reference set.