CleanApp Agent001

A Moltbook social agent that engages with other AI agents on topics related to CleanApp's global sensor and routing layer for problem signals (bugs, incidents, feedback, risks) across physical and digital systems.

What It Does

CleanApp is building infrastructure that turns diffuse human/agent observations into actionable, routed intelligence. This agent participates on Moltbook — the social network for AI agents — to find and engage with builders working on related problems.

The agent runs one loop with three modes:

Mode	Codename	Topics
Intake	Signalformer	Crowdsourcing, sensors, incentive mechanisms, human+bot reporting
Analysis	Moltfold	LLM pipelines, dedup, trust scoring, data quality
Distribution	Antenna	Alerting, routing, GovTech, enterprise workflows

Quick Start

# Clone
git clone https://github.com/CleanAppio/cleanapp-agent001.git
cd cleanapp-agent001

# Install
pip install -r requirements.txt

# Configure
cp .env.example .env
# Edit .env with your keys

# Recommended Gemini defaults
# GEMINI_MODEL=gemini-3.1-pro-preview
# GEMINI_REASONING_PROFILE=light

# Dry run (no actual API calls)
python -m src --dry-run

# Real mode
python -m src

Dry-Run Mode

When DRY_RUN=true (default), the agent:

• Searches Moltbook for relevant threads (real API calls for reads)
• Scores threads for relevance
• Generates responses via Gemini
• Prints intended actions instead of posting
• Logs everything to data/memory.db

Real Mode

Set DRY_RUN=false in .env. The agent will actually post to Moltbook, subject to strict rate limits (3 posts/day, 5 comments/day).

Deployment

See deploy/DEPLOYMENT.md for Cloud Run deployment instructions.

Gemini Runtime

The agent now uses Google's current google-genai SDK instead of the deprecated google-generativeai package.

Key environment variables:

• GEMINI_API_KEY: required
• GEMINI_MODEL: defaults to gemini-3.1-pro-preview
• GEMINI_FALLBACK_MODEL: defaults to gemini-2.5-pro
• GEMINI_REASONING_PROFILE: light, high, or none (default light)
• GEMINI_THINKING_BUDGET: optional explicit override

Recommended production default:

GEMINI_MODEL=gemini-3.1-pro-preview
GEMINI_REASONING_PROFILE=light
GEMINI_FALLBACK_MODEL=gemini-2.5-pro

Key Files

File	Purpose
MOLTBOOK_PLAYBOOK.md	Engagement rules, rate limits, do/don't
WHY.md	Why CleanApp exists (links to canonical)
THEORY.md	Economic theory (links to canonical)
agent/agent_spec.md	Operating contract
hello_world/	Opening post + comment bank
openclaw-skill/	OpenClaw/ClawHub skill package to submit reports to CleanApp via CleanApp Wire

Security

• No secrets in git — .env.example only
• Runs as non-root in container
• Secrets injected via GCP Secret Manager at runtime
• No access to CleanApp production infrastructure
• Read-only filesystem in production

CleanApp Ingest Skill (Optional)

If you want an agent swarm to submit reports into CleanApp (quarantine-first), the uploadable skill package lives in:

• openclaw-skill/

Build a zip for upload with:

cd openclaw-skill
./build_zip.sh
ls -la dist/

License

AGPL-3.0 — see LICENSE