القائمة
Screen recording + AI vision analysis via direct Gemini REST API. No browser, no Colab MCP, no CDP. Pure REST with 6-model fallback chain for unlimited vision calls.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Unified plan-and-execute workflow. Check whether a viable plan already exists, create and review one if needed, then execute task by task until the done criteria pass. Synthesizes the strongest parts of omoc-plan-swarm, biometrics-plan, and biometrics-work.
OMOC Plan Swarm -- activated when any agent receives a planning request. Triggers: plan, erstelle einen plan, create a plan, planung, planning, roadmap, strategy, strategie, konzept, blueprint. Instead of planning alone, the agent runs a 3-stage pipeline: parallel research, plan synthesis, critical review. Simple, small processes, fail-fast gates. Never plan alone.
| name | sin-vision-colab |
| description | Screen recording + AI vision analysis via direct Gemini REST API. No browser, no Colab MCP, no CDP. Pure REST with 6-model fallback chain for unlimited vision calls. |
| license | Apache-2.0 |
| compatibility | opencode |
| metadata | {"audience":"all-agents","workflow":"screen-vision-analysis","trigger":"screen-record, screenshot, vision-analyze, look-screen","version":"v4-gemini-rest-fallback"} |
Screen recording + AI vision analysis via direct Gemini REST API with a 6-model fallback chain.
AIzaSyA1THIC_... is revoked[Agent] → look-screen CLI → [Gemini REST API] → Vision Analysis
↓ ↑
screencapture -x base64 inline_data
↓ ↑
/tmp/screen.png 6-model fallback chain
AIzaSyCnRoGEoQJBAVssEu6BP1ojSBzIwV5r8_o| Priority | Model | Free Tier Quota | Purpose |
|---|---|---|---|
| 1 | gemini-2.5-flash | Unlimited RPM, 1M context | Primary vision |
| 2 | gemini-3-flash-preview | 5 RPM, 250K tokens | Backup vision |
| 3 | gemini-3.1-flash-lite-preview | 15 RPM, 250K tokens | Lightweight backup |
| 4 | gemma-3-27b | 30 RPM, 15K context | Fallback vision |
| 5 | gemma-3-12b | 30 RPM, 15K context | Fallback vision |
| 6 | gemma-3-4b | 30 RPM, 15K context | Last resort |
| Trigger | Aktion |
|---|---|
| "nimm bildschirm auf" | look-screen --record |
| "analysiere bildschirm" | look-screen --screenshot /tmp/screen.png --describe |
| "was siehst du?" | look-screen --once --screenshot /tmp/screen.png |
| "überwache bildschirm" | look-screen --interval 3 |
| "stop aufnahme" | look-screen --stop |
# Symlink: ~/.local/bin/look-screen → ~/.open-auth-rotator/tools/look_screen.py
look-screen --version
# Output: look-screen v4.0.0 (Gemini REST API + Fallback Chain)
Der API Key ist hardcoded im Script. Überschreiben via Environment Variable:
export GEMINI_VISION_API_KEY="your-key-here"
look-screen --status
# Expected:
# [look-screen] Vision Architecture: v4-gemini-rest-fallback
# [look-screen] Models (6): gemini-2.5-flash, gemini-3-flash-preview, ...
# [look-screen] Browser Automation: NONE (Pure REST)
# [look-screen] Status: Active
# Status prüfen
look-screen --status
# Screenshot analysieren
look-screen --screenshot /tmp/screen.png --describe
# Mit custom Prompt
look-screen --screenshot /tmp/screen.png --describe --prompt "Ist ein Fehler sichtbar?"
# Kontinuierliche Überwachung
look-screen --interval 3
# Screenshot + Analyse in einem Befehl
screencapture -x /tmp/screen.png && look-screen --screenshot /tmp/screen.png --describe
import subprocess
# 1. Screenshot machen
subprocess.run(["screencapture", "-x", "/tmp/screen.png"])
# 2. Analyse anfordern
result = subprocess.run(
["look-screen", "--screenshot", "/tmp/screen.png", "--describe"],
capture_output=True, text=True
)
analysis = result.stdout
NIEMALS versuchen, Vision-Analyse über Browser, Colab MCP oder CDP zu machen.
Immer look-screen CLI nutzen — das ruft direkt die Gemini REST API auf.