jiruidai/dsh-meta-orchestrator
A model-native meta-agent plugin for DeepSeek Harness that uses the underlying model’s reasoning and planning
Teaches the agent to synthesize a task-specific workflow at runtime instead of hard-coding pipelines or fixed topologies: analyze the request (ask if genuinely ambiguous), pick one of five proven patterns (prompt-chaining, parallel-workers, router, supervisor, evaluation-loop — each playbook loads only when picked), write the plan down via the orchestrate tool with stages, delegated roles, and verifiable success criteria, and validate + save it durably. The plan the agent follows becomes explicit, versioned, and durable rather than implicit.
Install
dsh plugin --profile web add dsh-meta-orchestratornpm package dsh-meta-orchestrator 0.2.0 (registry-verified 2026-08-26; 46/46 tests passing, live web profile verified 2026-08-14). dsh plugin --profile web add dsh-meta-orchestrator. DSH 0.1.0-rc.6 verified. MIT.
Compatibility
DeepSeek Harness 0.1.0-rc.6. Works for multi-step work where the agent's plan should be explicit, versioned, and durable.
Details
- Repo: jiruidai/dsh-meta-orchestrator
- Category: Web UI Enhancements
- Stars: 3
- Version: npm dsh-meta-orchestrator 0.2.0 (registry-verified 2026-08-26)
- Last push: 2026-08-14
- First seen: 2026-08-13
Recent updates
v0.2.0: five workflow patterns; runtime orchestrate tool; durable plan storage; ambiguity-first analysis.
FAQ
- How does the agent decide the workflow?
- It analyzes the request, picks one of five patterns (prompt-chaining, parallel-workers, router, supervisor, evaluation-loop), and writes the plan down with success criteria.
- Does it ask before acting?
- If the request is genuinely ambiguous, the agent asks before doing anything.
- Is the plan durable?
- Yes — orchestrate validates the structure and saves it durably, so the workflow is versioned and reviewable.
Alternatives
ztl34245881-commits/dsh-task-planner · lhmd/dsh-director-toolkit · omdsh-dev/dsh-advisor