mcpField
D
duoduo-pipeline

duoduo-pipeline

github.com/openduoAgent workflows1090 installsUpdated 4d ago
Repository

About duoduo-pipeline

Duoduo Pipeline is a pattern for pairing scheduled mechanical scripts with a dormant-by-default duoduo brain job. It filters incoming events before waking the LLM, so only signals that meet a threshold reach the main session.

Key features

  1. 01Polls sources with shell or Python scripts
  2. 02Notifies the brain job when new data appears
  3. 03Keeps the brain job dormant between events
  4. 04Filters signals against a threshold
  5. 05Sends actionable briefs to the main session

Use cases

  1. 01Monitor feeds or accounts for new posts
  2. 02Alert when prices or metrics cross thresholds
  3. 03Detect file or repository changes
  4. 04Relay webhook payloads to a brain job

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/openduo/duoduo --skill duoduo-pipeline

After installing, type/duoduo-pipelinein Claude Code to confirm it loaded.

Bundled filesreferences/notify-routing.mdreferences/pattern-mechanical-brain.md

Details

From the project README

Duoduo Pipeline

Connect a zero-cost mechanical layer to a duoduo brain job so the LLM only runs when there is something worth acting on. During idle periods the brain job stays dormant — no polling, no token spend.

The three-layer model

Mechanical layer  (shell / Python script — zero LLM)
      │
      │  duoduo session notify  ← fires only when new data arrives
      ▼
Brain job  (keepalive — dormant by default)
      │
      │  Notify  ← fires only when signal meets your threshold
      ▼
Main session  (your foreground chat)
  • Mechanical layer — a plain script scheduled by launchd, cron, or a systemd timer. It polls the source, compares against last-seen state, and calls duoduo session notify only when something new is detected. No LLM involved at any point.
  • Brain job — a keepalive duoduo job. It wakes on notify, processes the incoming data, decides whether the result is worth surfacing, and sends a Notify to your main session only when the signal clears your threshold. Then it goes dormant again.
  • Main session — your regular foreground chat. It receives a single, pre-filtered brief only when the brain has found something actionable.

Typical use cases

  • Feed / account monitoring — check a public feed every few minutes; wake the brain only on new posts.
  • Price or metric alerts — poll an API for a number; wake the brain only when it crosses a threshold.
  • File or repo change detection — watch a directory or git remote; wake the brain only on new commits or file changes.
  • Webhook relay — a lightweight receiver script forwards payloads to the brain via notify, keeping the brain dormant between events.
  • Data-source health watch — detect when an upstream source goes stale and surface an alert without running a full LLM cycle just to check.

References

Comments

Markdown supported