february2015/dsh-taskswarm
DSH port of TaskPlane: dependency-ordered waves run in parallel git-worktree lanes, with task packets, cross-model review, and crash recovery.
dsh-taskswarm (TaskSwarm 蜂群) is multi-agent task orchestration: it arranges a batch of tasks into dependency-ordered waves from the task DAG, runs multiple AI workers in parallel lanes isolated by git worktrees, then automatically reviews and merges their output. Each task is a packet of PROMPT.md (mission/steps/constraints) + STATUS.md (progress) for durable memory across context resets; checkpoint discipline auto-commits at step boundaries so a crashed worker never loses committed work; an independent reviewer scores each task against its Review Level (PASS merges, REVISE sends back); a file mailbox lets workers and the supervisor communicate asynchronously; and a conversational supervisor shares your session with start/pause/abort/integrate commands, bilingual 中文/English notifications auto-detected from session language.
Install
dsh plugin --profile web add https://github.com/february2015/dsh-taskswarm.gitGit-URL install per README (bilingual EN/zh; EN primary). npm dsh-taskswarm 404 verified 2026-09-03 — the README's dated 'back on Aug 19' notice is stale, so the documented git route is primary: dsh plugin --profile web add https://github.com/february2015/dsh-taskswarm.git (or clone + npm install && npm run build && dsh plugin --profile web add $(pwd)).
Compatibility
DSH web profile; git worktrees for lane isolation; Node per DSH requirements; upstream TaskPlane (Pi ecosystem) native port.
Details
- Repo: february2015/dsh-taskswarm
- Category: Workflow & Automation
- Stars: 0
- Version: git https://github.com/february2015/dsh-taskswarm.git (npm 404 verified)
- Last push: 2026-08-16
- First seen: 2026-08-14
Recent updates
README carries a dated install notice (Aug 17–18, 2026) about an npm unpublish cool-down; npm remains unpublished at enrichment time.
FAQ
- How are parallel tasks isolated?
- Every task runs in its own git worktree lane, with results merged into the taskswarm/orch integration branch after review.
- What survives a crashed worker?
- Checkpoint discipline auto-commits at step boundaries and each task keeps a PROMPT.md + STATUS.md packet, so committed work is never lost across context resets or crashes.
- How is quality controlled?
- An independent reviewer scores each task against its Review Level: PASS merges into the integration branch, REVISE sends it back for revision.
Alternatives
apheli0os/deepseek-harness-orchestrate · wellorbetter/dsh-product-delivery-workflow