Thomashighbaugh/orchestrate-router

Orchestrate Router

Routes amorphous user requests to the correct hub subcommand. When a user enters something like /orchestrate build me a login system or /ideation think about database tradeoffs, this skill classifies their intent across multiple dimensions and recommends the best subcommand.

When to Use

• User enters /orchestrate [text about what to implement] — route to the right execution pattern
• User enters /ideation [text about what to think about] — route to the right planning methodology
• User enters /harvest-context [text about what to save/fetch] — route to the right extraction method
• The /cc10x subcommand also acts as an intent-detecting router for BUILD/DEBUG/REVIEW/PLAN

Scripts

| Script | Hub | Purpose | |--------|-----|---------| | scripts/route-orchestrate.mjs | /orchestrate | Routes to 27 execution patterns | | scripts/route-ideation.mjs | /ideation | Routes to 21 planning/research methods | | scripts/route-harvest.mjs | /harvest-context | Routes to 13 extraction/context methods + auto-vectorize |

How Routing Works

Each subcommand has a profile scored across multiple dimensions:

| Dimension | What It Measures | |-----------|-----------------| | Task type | build, fix, refactor, analyze | | Autonomy | guided, periodic, fire-and-forget | | Complexity | simple, moderate, complex, multi | | Confidence | clear, vague, bug report | | Scale | solo, small team, swarm | | Quality | draft, production, critical | | Time | asap, balanced, thorough | | Keywords | Strong signals (e.g., "fire and forget" → ralph) | | Anti-keywords | Negative signals (e.g., "research" → not build patterns) |

The classifier extracts signals from the user's natural language and scores each subcommand. The highest-scoring match is recommended with a confidence percentage.

Orchestrate Router Examples

| User says | Routes to | Why | |-----------|-----------|-----| | build me a login system | autopilot | Full autonomy from idea to code | | fix this TypeScript error | deep | Root-cause trace + requirements | | research auth tradeoffs | ccg | Multi-model synthesis | | CI build is broken | remediate | Auto-fix CI failures | | prototype an app idea | vibe-code | Conversational rapid prototyping | | ship the feature end to end | gsd | Full pipeline to PR | | complex multi-file with quality | swarm | Gated QA pipeline | | keep working on this until it passes | ralph | Persistent loop | | refactor the utils module | evolutionary | Incremental with fitness | | add a feature to existing codebase | brownfield | Integration-aware |

Ideation Router Examples

| User says | Routes to | Why | |-----------|-----------|-----| | plan the auth system | plan | Goals → ordered tasks | | brainstorm landing page ideas | brainstorm | Free divergence | | vague idea about notifications | deep | Socratic interview | | compare Postgres vs SQLite | research | Multi-model synthesis | | model the payment domain | ddd | Bounded contexts, aggregates | | map system dependencies | graph | Visual relationship mapping | | stress-test this API design | adversarial-debate | Oppositional critique | | set coding standards | constitution | Governance principles | | break down into small tasks | plan | Ordered with acceptance criteria |

Harvest Context Router Examples

| User says | Routes to | Why | |-----------|-----------|-----| | save what we learned | session | Extract session knowledge | | document the codebase | codebase | Hierarchical AGENTS.md | | create a reusable workflow | skill | SKILL.md from pattern | | fetch React hooks docs | docs | Context7 MCP lookup | | ingest this PDF | consume | File/URL → context | | compress to save tokens | compress | 4-layer compression | | set up a knowledge base | secondbrain | Local markdown+Git | | save this as a convention | rule | Project rule |

Auto-Vectorization (Harvest Context Only)

The harvest router integrates with the vectorize-context skill. After routing to any subcommand with writes: true (session, codebase, memory, consume, secondbrain, journal), the router automatically calls ensureIndexed() from veclib.mjs to refresh the vector DB.

This means:

• A user harvests session knowledge → context files written → vector DB auto-updated → next semantic query sees new content
• No manual triggering needed
• If no files changed, ensureIndexed() returns instantly (no model load)
• If files changed, only stale files are re-indexed

Usage

# Route an orchestration request
node {skill_dir}/scripts/route-orchestrate.mjs "build me a login system"

# Route an ideation request
node {skill_dir}/scripts/route-ideation.mjs "compare database options"

# Route a harvest request (auto-vectorizes on write subcommands)
node {skill_dir}/scripts/route-harvest.mjs "save what we learned"

# Environment variable for all three
REQUEST="fix the broken CI pipeline" node {skill_dir}/scripts/route-orchestrate.mjs

Output Format

All routers output JSON on stdout with the same shape:

{
  "request": "build me a login system",
  "recommended": "autopilot",
  "confidence": 26,
  "reason": "Full autonomy — idea to working code",
  "description": "Best for: building something from scratch...",
  "autoVectorize": false,
  "ranked": [
    { "subcommand": "autopilot", "score": 26 },
    { "subcommand": "vibe-code", "score": 21 },
    { "subcommand": "plan-execute", "score": 15 }
  ]
}

The autoVectorize field is only present in the harvest router output.

Integration

To wire into a hub subcommand (/orchestrate, /ideation, /harvest-context):

1. When user enters /orchestrate [text] without a recognized subcommand name:
   • Run route-orchestrate.mjs with the text
   • Take recommended from output
   • Route to that subcommand via hubMenu(action: "route", hub: "orchestrate", subcommand: recommended)
2. Same pattern for /ideation and /harvest-context
3. For harvest with autoVectorize: true, call ensureIndexed() after the extraction/save completes

Dependencies

• Node.js 18+
• vectorize-context skill (optional, for auto-vectorize in harvest router)