dispatch_work_order

Benchmarked against: Anthropic — Code execution tool MCP Tool: dispatch_work_order Pattern: Fire-and-forget background execution

The Dispatch Engine executes Work Orders using configurable AI engines. It runs as a background process — submit the WO, check results later.

---

Usage

dispatch_work_order(
    work_order_id="WO-2026-0215-007",
    engine="groq"    # or gemini, deepseek, claude, etc.
)

Parameters

Parameter	Type	Default	Description
work_order_id	string	required	The WO ID to execute
engine	string	"claude"	Execution engine (see table below)

Available engines

Engine	Cost	Best for
ingest	Free	Batch file ingest into UB. Put file paths in WO description.
groq	Free	Research, analysis, simple text tasks
groq-search	Free	Groq + built-in web search (intel/research)
gemini	Cheap	General tasks, good quality
gemini-search	~$0.014	Gemini + Google Search (most authoritative)
deepseek	Very cheap	Analysis, reasoning tasks
mistral	Cheap	European model, alternative option
zhipu	Cheap	Chinese NLP, agent tool-calling (GLM-5)
claude	~$1-2/run	ONLY for tasks requiring code editing/file operations

Engine selection principle

CP value is NOT "cheapest possible" but "minimum cost that gets the job done RIGHT."

• Trivial tasks → groq (free)
• Important research → gemini (cheap but quality)
• Code/file operations → claude (expensive but necessary)

Event-driven dispatch

Only call dispatch_work_order when Captain or 小西 explicitly decides to execute. Do not call proactively — wait for human instruction.

The worker runs as a background process (fire-and-forget). Use check_dispatch() to see results later.

Known limitations

• groq cannot do file system operations (cp, mv, mkdir). Use claude engine for file ops.
• groq may hallucinate version numbers. Verify results for important intel.
• claude engine is expensive (~$1-2 per run). Only use when necessary.