dustcraw

Airborne signal absorber — collects floating public data (RSS/API/web) and feeds the AI ecosystem.

Core Principle: Revenue & User Value First

Every feature and data source is prioritized by:

1. Profitability — Does it contribute to revenue?
2. Desirability — Do users love this content?
3. Demand — Is there a real need for this information?
4. Usage — Is the source high-traffic/engagement?

Architecture

• Patch Model: TTL-based configuration adjustments via resolveConfig().
• DecisionGate: Autonomous multi-AI deliberation for self-amplifying system changes.
• AdaptiveGovernor: Dynamic thresholding based on promotion trends.
• Handoff System: Post-deliberation task routing to executor agents.
• SignalStore (v0.2.0): Local JSONL-based storage for promoted signals, ensuring durability even when the brain is unavailable.
• Brain Isolation Strategy (v0.2.0): Decoupled architecture using BrainClientAdapter. dustcraw no longer requires a direct brain connection for its core loop, promoting portability and fault tolerance.

New Feature: Self-Amplifying Architecture

As of v0.2.0, dustcraw implements a self-amplifying feedback loop:

• Adaptive Budgeting: The system fetches user interaction stats from the brain to dynamically adjust PromotionBudget. High-value signals lead to increased ingestion capacity.
• Interest Drift Detection: Automatically identifies shifts in user interest and adjusts InterestTermStore via autonomous deliberation.

New Feature: Autonomous Deliberation Handoff

As of v0.1.1, dustcraw supports the Autonomous Deliberation Handoff pattern:

• Synthesis Consumption: Automatically polls mcp-deliberation state files for completed deliberations with auto_execute: true.
• Structured Tasks: Parses actionable_tasks (id, task, files, priority) to create precise implementation jobs.
• Shared Envelope: Uses the ecosystem-wide SharedEnvelope format for cross-agent communication.
• Inbox Watcher: A background daemon (bin/inbox-watcher.mjs) that picks up tasks and executes them via specialized AI skills (autopilot, TDD, parallel execution).

Setup & Usage

npm install
npm run build
dustcraw init   # Create default config
dustcraw start  # Run the collection daemon

Tech Stack

• TypeScript (ESM)
• Vitest
• Node.js
• systemd (Deployment)