By sabirmgd
Multi-agent swarm intelligence engine. Simulate how diverse stakeholders react to scenarios and predict outcomes using knowledge graphs, agent personas, and interactive dashboards.
Extract entities and relationships from seed content for SwarmFish simulations. Use when building a knowledge graph from text.
Conduct in-character interviews with SwarmFish simulation agents. Responds as the agent based on their persona and simulation history.
Generate detailed agent personas from entities for SwarmFish simulations. Creates MBTI, background, goals, biases, and system prompts.
Generate comprehensive prediction reports from SwarmFish simulation data. Synthesizes agent actions, sentiment evolution, and graph metrics into evidence-based predictions.
Execute individual agent turns in SwarmFish simulation rounds. Stays in character and produces actions based on persona and context.
This plugin requires configuration values that are prompted when the plugin is enabled. Sensitive values are stored in your system keychain.
Uses power tools
Uses Bash, Write, or Edit tools
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
default_sim_modelDefault model for simulation agent turns (opus, sonnet, haiku)
${user_config.default_sim_model}auto_open_dashboardAutomatically open dashboard in browser after generation (true/false)
${user_config.auto_open_dashboard}default_report_modelDefault model for report generation (opus, sonnet, haiku)
${user_config.default_report_model}Multi-agent swarm intelligence engine for Claude Code.
Predict outcomes by simulating how diverse stakeholders interact. Feed in a topic, and SwarmFish builds a knowledge graph, generates agent personas, runs a multi-round simulation, and produces a prediction report with an interactive HTML dashboard.
Inspired by MiroFish — reimagined as a Claude Code plugin with file-based knowledge graphs and zero external services.
# Add the marketplace
/plugin marketplace add sabirmgd/swarmfish
# Install the plugin
/plugin install swarmfish@swarmfish
# Install optional tools (graphviz + networkx)
/swarmfish:swarm setup
Or test locally:
git clone https://github.com/sabirmgd/swarmfish.git
claude --plugin-dir ./swarmfish
# 1. Initialize a simulation from a topic
/swarmfish:swarm init "How will the developer community react to a new open-source AI model release?"
# 2. Run the simulation
/swarmfish:swarm run
# 3. Generate prediction report
/swarmfish:swarm report
# 4. Open visual dashboard
/swarmfish:swarm dashboard
# 5. Chat with any agent
/swarmfish:swarm chat "OpenAI"
Topic/Document
│
▼
┌─────────────┐ ┌──────────────┐ ┌───────────────┐
│ Extract │────▶│ Generate │────▶│ Simulate │
│ Entities & │ │ Agent │ │ Rounds │
│ Relationships│ │ Personas │ │ (parallel) │
└─────────────┘ └──────────────┘ └───────┬───────┘
│
┌──────────────┐ ┌────────▼────────┐
│ Generate │◀────│ Update Graph │
│ Report │ │ Memory │
└──────┬───────┘ └─────────────────┘
│
┌──────▼───────┐
│ Dashboard │
│ (HTML/D3) │
└──────────────┘
| Command | Description |
|---|---|
init <topic> | Seed a new simulation from a topic or scenario file |
run [sim-id] | Execute simulation rounds with agent interactions |
report [sim-id] | Generate a comprehensive prediction report |
dashboard [sim-id] | Build and open HTML visualization dashboard |
chat <agent> | Interview a specific agent in character |
graph [sim-id] | Inspect and query the knowledge graph |
metrics [sim-id] | Run NetworkX graph metrics (PageRank, centrality) |
svg [sim-id] | Generate force-directed SVG graph |
pdf [sim-id] | Export report as professional PDF |
query <name> | Query actions with jq (stats, sentiment, by-agent...) |
status [sim-id] | Show simulation status or list all |
inject <event> | Inject a God's Eye event into next round |
check | Verify tools and dependencies |
setup | Install optional dependencies |
config | Show and validate configuration |
On first use, SwarmFish creates .swarmfish/config.yaml in your project. Edit it to customize:
# Which Claude model for each phase
models:
entity_extraction: "sonnet"
persona_generation: "sonnet"
simulation: "haiku" # Fast for many agent turns
report: "opus" # Best quality for synthesis
# Simulation settings
simulation:
max_rounds: 10
max_agents: 15
parallel_agents: 3 # How many agents run in parallel
# Dashboard
dashboard:
auto_open: true
theme: "dark"
See config/default-config.yaml for all options.
Create reusable scenarios as YAML files:
name: "Market Reaction Analysis"
seed:
type: "text"
content: |
Breaking: Company X announces...
prediction_question: |
How will stakeholders react?
interventions:
- round: 5
event: "Breaking: New regulation announced..."
See scenarios/example.yaml for a full example.
The HTML dashboard (auto-generated, opens in browser) includes:
| Tab | Visualization |
|---|---|
| Graph | D3.js force-directed entity relationship graph |
| Simulation | Round-by-round action timeline with agent cards |
| Agents | Persona cards with stats, MBTI, activity bars |
| Sentiment | Chart.js line chart of opinion evolution |
| Report | Rendered prediction report |
| Chat | Interview conversation history |
| Tool | Install | What it adds |
|---|---|---|
graphviz | brew install graphviz | Force-directed SVG graph layouts |
networkx | pip install networkx numpy | PageRank, centrality, community detection |
md-to-pdf | npm i -g md-to-pdf | Professional PDF export |
jq | brew install jq | CLI action queries |
Run /swarmfish:swarm setup to install all optional tools interactively.
npx claudepluginhub sabirmgd/swarmfish --plugin swarmfishTurn websites and browser workflows into deterministic API scripts with JSON and CSV outputs.
A decision operating system for high-stakes choices. Simulates disagreement across 5 independent personas, stress-tests assumptions, and converges on what actually holds up.
Commands for scenario simulation and decision modeling
Strategy & stakeholder skills: Competitor Signal Tracker, Competitive Intelligence Monitor, Stakeholder Influence Mapper, Strategic Narrative Generator, Executive Update, Ambiguity Resolver.
Expert panel discussion system with 22 curated personas. Includes 2 agents, 1 command, and 22 expert definitions.
Three lenses for clarity: vague requirements → specs (vague), strategy blind spots → 4-quadrant playbook (unknown), content vs form → leverage shift (metamedium)
Compile humans into AI agents through deep interviews. Conducts comprehensive behavioral interviews, analyzes work artifacts via MCP tools, and generates installable Claude Code plugins that embody a person's decision-making, communication style, and expertise.