PolyForge Plugin

Backend engineering command center for polyglot developers on OpenClaw

PolyForge is an OpenClaw CLI plugin that provides a structured, research-first, language-aware workflow for backend engineers working in Python, Rust, TypeScript and more.

Features

• 8 Specialized Agents — Atlas orchestrator, Sprint/Forge workers, Architect, Researcher, Analyst, Critic, Explorer
• 8 Workflow Definitions — brainstorm, triage, research, intake, review, plan, execute, work
• Language-Aware — Auto-detects project language, injects relevant skills
• Research-First Workflow — Evaluate technologies before building with structured tech briefs
• Task Intake — Turn raw PM requests and issues into actionable plans
• Anti-Hallucination Guardrails — 8 rules injected into every prompt
• Autorun Loop — Self-correcting execution with configurable iteration limits
• Todo Tracking — Persistent todo management across sessions
• Quality Gates — Configurable enforcement levels for outputs
• Adaptive Learning — Learns approval patterns over time

Installation

Via OpenClaw CLI (Recommended)

openclaw plugins install @dv.nghiem/polyforge
openclaw pf-setup

From Source

git clone https://github.com/DVNghiem/PolyForge.git
cd PolyForge/plugin
npm install
npm run build
npx pf-setup

Quick Start

# Switch to the Atlas orchestrator
/pf atlas

# Triage an incoming task
/triage PM wants async audit logging for all API calls

# Research a technology
/research axum vs actix-web for our HTTP layer

# Full work pipeline: intake → research → plan → execute → review
/work implement rate limiting middleware for FastAPI

# Check status
/pf status

# List all agents
/pf list

Commands

| Command | Description | |---------|-------------| | /pf [subcommand] | Main command — status, health, config, list, agent switch, off | | /triage [task] | Classify and route a task | | /research [topic] | Technology investigation workflow | | /intake [description] | Turn a raw task into a structured brief | | /plan [topic] | Produce an execution plan | | /execute [plan] | Run an approved plan | | /work [task] | Full pipeline: intake → plan → execute → review | | /review [target] | Code or plan review | | /brainstorm [topic] | Structured ideation with ≥3 approaches | | /todos | List all tracked todos | | /autorun [maxIter] | Start self-correcting execution loop | | /stop | Stop the autorun loop |

Workflows

PolyForge provides 8 predefined workflow definitions:

| Workflow | Phases | Use Case | |----------|--------|----------| | brainstorm | problem_framing → divergent → convergent → decision | Ideation with multiple approaches | | triage | classify → assess → route | Quick task classification | | research | scope → investigate → synthesize → present | Technology evaluation | | intake | parse → gap_analysis → research_trigger → plan_stub → handoff | Request intake | | review | scope_review → execute_review → recommend → signoff | Formal review process | | plan | context_gather → decompose → gap_review → document → approve | Planning | | execute | load_plan → resolve_deps → execute_batch → next_batch → update_plan | Plan execution | | work | intake → research → plan → execute → review → ship | Full development cycle |

Tools

The plugin registers these tools for use by agents:

| Tool | Description | |------|-------------| | pf_delegate | Delegate task to a specialized agent | | pf_spawn_acp | Spawn an autonomous coding process | | pf_research | Conduct technology research | | pf_intake | Process raw task into structured brief | | pf_search | Web search for technology comparison | | pf_checkpoint | Save/restore execution state | | pf_todo_create | Create a tracked todo | | pf_todo_list | List all todos | | pf_todo_update | Update todo status | | pf_phase_transition | Transition workflow phase | | pf_spawn_subagent | Spawn a subagent with specific role |

Configuration

Add to your OpenClaw plugin config:

{
  "polyforge": {
    "max_auto_iterations": 50,
    "preferred_language": "mixed",
    "research_depth": "standard",
    "checkpoint_dir": "workspace/checkpoints",
    "workflow_phase_enforcement": {
      "enabled": true,
      "default_approval_level": "medium"
    }
  }
}

Configuration Options

| Option | Type | Default | Description | |--------|------|---------|-------------| | max_auto_iterations | integer | 50 | Maximum autorun loop iterations | | preferred_language | string | "mixed" | Primary language (python, rust, typescript, mixed) | | research_depth | string | "standard" | Research depth (shallow, standard, deep) | | checkpoint_dir | string | "workspace/checkpoints" | Directory for execution checkpoints | | todo_enforcer_enabled | boolean | true | Enable todo continuation injection | | comment_checker_enabled | boolean | true | Enable AI slop comment detection | | team_agent_ids | array | [] | OpenCode agent IDs for team delegation |

Workflow Phase Enforcement

{
  "workflow_phase_enforcement": {
    "enabled": true,
    "default_approval_level": "medium",
    "per_workflow_overrides": {
      "review": "high",
      "triage": "low"
    }
  }
}

Intelligent Features

{
  "intelligent_features": {
    "adaptive_learning": {
      "enabled": true,
      "min_samples_for_auto_approve": 5,
      "auto_approve_threshold": 0.9
    },
    "proactive_suggestions": {
      "enabled": true,
      "suggestion_interval_ms": 30000
    },
    "quality_gates": {
      "enabled": true,
      "enforcement_level": "moderate",
      "default_threshold": 0.75
    }
  }
}

Agents

| Agent | Role | Tier | |-------|------|------| | Atlas | Work coordinator & orchestrator | Strategy | | Sprint | Quick implementation worker | Execution | | Forge | Deep specialist (complex tasks) | Execution | | Architect | System design (read-only) | Coordination | | Researcher | Technology evaluation | Coordination | | Analyst | Gap analysis & assessment | Coordination | | Critic | Code & plan reviewer | Coordination | | Explorer | Codebase search & navigation | Coordination |

License

MIT

Links

• Main Repository
• OpenClaw Documentation
• Issue Tracker