icetomoyo/dsh_workflow
Bring Claude Code's UltraCode mode to DSH, and upgrade DSH's one-time multi-Agent scheduling to a workflow layer that can be generated, saved, managed, observable, and recoverable
DSH Workflow upgrades DeepSeek Harness multi-agent execution into a reusable, governed, observable, and resumable workflow layer, implementing the full KodaX workflow model. Named workflows can be saved, discovered, and run by name. A durable run graph persists events, artifacts, result summaries, and per-run cost. Snapshot rerun and effect-cache resume handle interruptions without restarting from scratch. Manifest-level preflight checks enforce token reservations, concurrency caps, and agent limits. The native DSH workflow tool handles single foreground fan-out; this plugin is the reusable process layer above it.
Install
dsh plugin --profile web add "github:icetomoyo/dsh_workflow#main"Compatibility
DSH 0.0.1-rc.2+, Node.js 22.19+. MIT license.
Details
- Repo: icetomoyo/dsh_workflow
- Category: Coding & Development
- Stars: 55
- Version: Latest from main branch
- Last push: 2026-09-05
- First seen: 2026-08-12
FAQ
- What is the difference from the native DSH workflow tool?
- The native DSH workflow tool is best for a one-time parallel fan-out. DSH Workflow adds the reusable layer: named workflows, persistent run graphs, snapshot resume, cost tracking, and governance. The two coexist.
- Can workflows be saved and reused across sessions?
- Yes. Workflows are named and stored; run by name with /workflow parallel-investigation or run_workflow. Results, events, and costs persist in a durable run graph.
- What happens when a workflow is interrupted mid-run?
- Use snapshot rerun to restart from the last checkpoint, or effect-cache resume to continue only the unfinished portions — completed steps are not re-run.
Alternatives
btspoony/mstar-harness · omdsh-dev/dsh_workflow