project-task-router

Project Task Router

Live Router Demo

Cross-platform task triage plugin for Codex, Claude Code, and Copilot CLI.

It routes incoming work requests to:

• A top domain route
• A recommended skill stack (primary + secondary)
• A model profile (deep_reasoning, balanced, fast_execution)
• Clarification mode when intent is ambiguous

Features

• Deterministic weighted router script: skills/project-task-router/scripts/route_task.py
• 8-domain routing taxonomy
• Token/phrase/regex signal scoring with negative-signal penalties
• Ambiguity handling with alternatives
• Dynamic installed-skill discovery with alias fallback
• Missing-capability detection (missing_capabilities)
• Subagent orchestration pattern for review then remediation

Domain Taxonomy

1. market-discovery-validation
2. product-strategy-positioning
3. architecture-build-planning
4. feature-implementation
5. ui-ux-frontend
6. security-compliance-reliability
7. data-analytics-ai
8. platform-ops-reliability

Requirements

• Python 3.9+
• Optional installed skill ecosystem:
  • nmalinowski/groundwork
  • nmalinowski/agents
  • nmalinowski/ui-ux-pro-max-skill
  • 21st.dev/community/components
  • nano-banana-pro

CLI Usage

Basic routing:

python skills/project-task-router/scripts/route_task.py --task "Implement HIPAA-compliant audit logging"

Markdown output:

python skills/project-task-router/scripts/route_task.py --task "Plan API architecture" --format markdown

Route with explicit installed-skill list:

python skills/project-task-router/scripts/route_task.py --task "Improve frontend layout" --installed-skills "ui-ux-pro-max,design-system,product-design"

Route with installed-skill file:

python skills/project-task-router/scripts/route_task.py --task "Build RAG pipeline" --installed-skills-file installed-skills.json

Skill Discovery Order

When --installed-skills and --installed-skills-file are not provided, discovery runs in this order:

1. Environment variables: ROUTER_INSTALLED_SKILLS, INSTALLED_SKILLS, CODEX_INSTALLED_SKILLS, COPILOT_INSTALLED_SKILLS, CLAUDE_INSTALLED_SKILLS
2. Relative manifests:
   • installed-skills.json
   • .codex/installed-skills.json
   • .claude/installed-skills.json
   • copilot/installed-skills.json
3. Home manifests:
   • ~/.codex/installed-skills.json
   • ~/.claude/installed-skills.json
4. Plugin-local manifests:
   • <plugin-root>/installed-skills.json
   • <plugin-root>/.codex/installed-skills.json
   • <plugin-root>/.claude/installed-skills.json
   • <script-dir>/installed-skills.json

First existing manifest wins.

Output Fields

• route: selected domain or needs-clarification
• confidence: low, medium, high
• primary_skills and secondary_skills
• alternatives: close competing routes
• missing_capabilities: required capability IDs not resolved from known inventory
• evidence: domain score map
• note: routing/discovery hints

Subagent Workflow

Use this sequence when executing routed work:

1. Run router and confirm route/model/skills.
2. Spawn review subagent for comprehensive findings (correctness, regressions, security, tests).
3. After review completes, spawn remediation subagent.
4. Remediation policy:
   • Auto-fix low, medium, high
   • Escalate critical findings for explicit approval

Development

Run tests:

python -m unittest discover -s tests -v

License

MIT (see plugin.json).