Back to Skills

coordinate-swarm

pjt222
Updated 2 days ago
6 views
17
2
17
View on GitHub
Metaautomationdesign

About

The coordinate-swarm skill provides patterns like stigmergy and quorum sensing for building distributed systems that self-coordinate without central control. It helps developers design resilient, event-driven architectures by focusing on signal design, agent autonomy, and tuning emergent behavior. Use it when you need to eliminate coordination bottlenecks or replace fragile orchestration with decentralized coordination.

Quick Install

Claude Code

Recommended
Primary
npx skills add pjt222/agent-almanac -a claude-code
Plugin CommandAlternative
/plugin add https://github.com/pjt222/agent-almanac
Git CloneAlternative
git clone https://github.com/pjt222/agent-almanac.git ~/.claude/skills/coordinate-swarm

Copy and paste this command in Claude Code to install this skill

Documentation

Coordinate Swarm

Stigmergy + local rules + quorum → coherent collective, no central ctrl.

Use When

  • Distributed sys → no central bottleneck
  • Self-coord teams → no mgr overhead
  • Event-driven arch → shared state, not direct msg
  • Works @3 agents → breaks @30 → scale
  • Bootstrap new swarm domain (forage-resources, build-consensus)
  • Replace fragile central orch → resilient emergent

In

  • Required: Agents desc (workers, services, team)
  • Required: Collective goal / target behavior
  • Optional: Current coord + fail modes
  • Optional: Agent count → pattern choice
  • Optional: Latency tolerance (realtime vs eventual)
  • Optional: Env constraints (shared state, bandwidth)

Do

Step 1: Classify Problem

  1. Map: who agents, what do, where coord breaks
  2. Classify:
    • Foraging → search distributed res (forage-resources)
    • Consensus → agree collective decision (build-consensus)
    • Construction → build shared structure
    • Defense → detect threats (defend-colony)
    • Division of labor → self-organize roles
  3. Fail mode:
    • Single point fail (central ctrl)
    • Comm bottleneck (too many msg)
    • Coherence loss (drift, no feedback)
    • Rigidity (no adapt)

→ Clear class + fail mode → pattern choice.

If err: no single class → composite → decompose. Heterogeneous → layered coord (homogeneous clusters + inter-cluster stigmergy).

Step 2: Design Signals

Indirect comm channels.

  1. Shared env (DB, queue, FS, board)
  2. Signal types:
    • Trail: accumulate on success paths (ant pheromone)
    • Threshold: counter → behavior switch
    • Inhibition: repel from exhausted areas
  3. Props:
    • Decay: fade rate → no stale dominance
    • Reinforce: success strengthens
    • Radius: propagation range
  4. Signal → behavior map:
    • Signal X > T → action A
    • A done → deposit Y
    • No signal → default explore
Signal Design Template:
┌──────────────┬───────────────────┬──────────────┬────────────────────┐
│ Signal Name  │ Deposited When    │ Decay Rate   │ Agent Response     │
├──────────────┼───────────────────┼──────────────┼────────────────────┤
│ success-trail│ Task completed OK │ 50% per hour │ Follow toward      │
│ busy-marker  │ Agent starts task │ On completion│ Avoid / pick other │
│ help-signal  │ Agent stuck >5min │ 25% per hour │ Assist if nearby   │
│ danger-flag  │ Error detected    │ 10% per hour │ Retreat & report   │
└──────────────┴───────────────────┴──────────────┴────────────────────┘

→ Signal table: deposit conds + decay + responses. Simple + composable.

If err: too complex → 2 signals (attract/repel). Add nuance after basic works.

Step 3: Local Rules

Simple rules, local info only.

  1. Perception radius (what sense?)
  2. 3-7 rules, priority order:
    • Rule 1 (safety): danger-flag → flee
    • Rule 2 (response): help-signal + idle → move toward
    • Rule 3 (exploit): success-trail → follow strongest
    • Rule 4 (explore): no signal → random + unexplored bias
    • Rule 5 (deposit): task done → deposit success-trail
  3. Each rule:
    • Local: only what agent perceives
    • Simple: one if-then
    • Stateless (pref): no past mem
  4. Mental test → does collective behavior emerge?

→ Prioritized rules, independent exec → target behavior emerges.

If err: no emergence → feedback loop needed. Add signal for collective state + adjust rule.

Step 4: Quorum Thresholds

Trigger collective changes when enough agree.

  1. Collective decisions:
    • Explore → exploit mode
    • New worksite commit / abandon
    • Normal → emergency
  2. Per decision:
    • Threshold: # / % agents agreeing
    • Window: signal count period
    • Hysteresis: different on/off thresh → no osc
  3. Quorum = signal accumulation:
    • Fav agent → vote-signal
    • Votes > thresh in window → activate
    • Votes < deact thresh → reverse

→ Leaderless decisions. Hysteresis gap → no rapid osc.

If err: oscillation → widen hyst gap (70/30). Never reaches quorum → lower thresh / widen window. Too slow → shrink window (beware premature).

Step 5: Test + Tune

  1. Pilot 5-10 agents
  2. Observe:
    • Converges on behavior?
    • How long?
    • Conditions change mid-task → what?
    • Agents fail / added → what?
  3. Tune params:
    • Decay: fast → no memory; slow → stale dominates
    • Quorum: low → premature; high → paralysis
    • Explore/exploit balance: too explore → inefficient; too exploit → local optima
  4. Stress:
    • Remove 30% agents → recover?
    • Double count → still coord?
    • Conflict signals → resolve / deadlock?

→ Tuned params, self-organizes, recovers, scales.

If err: stress fails → too tightly coupled. Simplify: fewer signals, faster decay, robust default. Swarm w/ zero-signal default > signal-dependent swarm.

Check

  • Problem classified (foraging / consensus / construction / defense / labor)
  • Signal table: deposit + decay + response
  • Rules simple + local + prioritized (3-7)
  • Quorum w/ hysteresis → no osc
  • Small test → emergent behavior matches goal
  • Stress test → graceful degradation

Traps

  • Signal bloat: Too many types → confusion. Start 2 (attract/repel)
  • Fake local: Rule needs global state → not local. Refactor
  • No decay: Fossilized coord state. Half-life per task scale
  • Zero hysteresis: Rapid osc. Deact < act always
  • Homogeneity assumed: Diff caps → role-diff rules (scale-colony)

  • forage-resources — res search + explore-exploit
  • build-consensus — distrib agreement deep-dive
  • defend-colony — collective defense on signal framework
  • scale-colony — scaling past initial coord
  • adapt-architecture — morphic arch transform
  • deploy-to-kubernetes — distrib sys deploy
  • plan-capacity — capacity + swarm scaling
  • coordinate-reasoning — AI self-variant; stigmergy → ctx mgmt

GitHub Repository

pjt222/agent-almanac
Path: i18n/caveman-ultra/skills/coordinate-swarm
0
agentsagentskillsai-assisted-developmentclaude-codeskillsteams

Related Skills

content-collections

Meta

This skill provides a production-tested setup for Content Collections, a TypeScript-first tool that transforms Markdown/MDX files into type-safe data collections with Zod validation. Use it when building blogs, documentation sites, or content-heavy Vite + React applications to ensure type safety and automatic content validation. It covers everything from Vite plugin configuration and MDX compilation to deployment optimization and schema validation.

View skill

polymarket

Meta

This skill enables developers to build applications with the Polymarket prediction markets platform, including API integration for trading and market data. It also provides real-time data streaming via WebSocket to monitor live trades and market activity. Use it for implementing trading strategies or creating tools that process live market updates.

View skill

creating-opencode-plugins

Meta

This skill helps developers create OpenCode plugins that hook into 25+ event types like commands, files, and LSP operations. It provides the plugin structure, event API specifications, and implementation patterns for JavaScript/TypeScript modules. Use it when you need to intercept, monitor, or extend the OpenCode AI assistant's lifecycle with custom event-driven logic.

View skill

sglang

Meta

SGLang is a high-performance LLM serving framework that specializes in fast, structured generation for JSON, regex, and agentic workflows using its RadixAttention prefix caching. It delivers significantly faster inference, especially for tasks with repeated prefixes, making it ideal for complex, structured outputs and multi-turn conversations. Choose SGLang over alternatives like vLLM when you need constrained decoding or are building applications with extensive prefix sharing.

View skill