KanoNoUta/dsh-captain
GPT plans dependency DAGs, DeepSeek workers execute tasks with adaptive parallelism, and an optional GPT reviewer audits incremental Git diffs and drives repair rounds.
Captain: a dual-face DSH plugin exposing one synthetic captain model route where GPT is the planner and independent reviewer while DeepSeek workers make incremental repository changes. The Host registers captain through the normal LLM adapter directory (model selector shows e.g. 'GPT-5.6 Terra -> DeepSeek V4 Flash'; route id captain:<planner>-><worker>, advertises text+image input and Balanced/High Quality/Ultra reasoning efforts). The planner returns a JSON dependency DAG; workers with disjoint file ownership run concurrently under a token budget and adaptive parallel limit; a reviewer receives acceptance criteria, worker reports, and the current incremental Git diff; a passing review advances an in-memory checkpoint.
Install
dsh plugin --profile web add github:KanoNoUta/dsh-captainGitHub install per README: dsh plugin --profile web add github:KanoNoUta/dsh-captain — the package ships its bundle manifest and built Host/Client entry points, so a Web profile installs it directly (cordis.patch.yml; no integration patch needed). npm package @kanonouta/dsh-captain not published as of 2026-08-26. Source development inside a DSH checkout uses git submodule + integration patch. Configure relay providers in normal LLM settings and reference their provider/model ids from the Captain settings namespace.
Compatibility
DeepSeek Harness web profile. OpenAI-compatible provider routes via the existing LLM adapters — no OAuth, no second credential store.
Details
- Repo: KanoNoUta/dsh-captain
- Category: Workflow & Automation
- Stars: 3
- Version: GitHub source (no npm package)
- Last push: 2026-08-15
- First seen: 2026-08-14
Recent updates
synthetic captain route; GPT planner + DeepSeek workers; JSON dependency DAG; concurrent workers; independent review + repair loop; reasoning efforts.
FAQ
- How do I use it?
- Install via dsh plugin --profile web add github:KanoNoUta/dsh-captain, configure relay providers in normal LLM settings, then pick a captain model route in the selector.
- Which models play which role?
- GPT is the planner and independent reviewer; DeepSeek workers make the incremental repository changes; route id is captain:<planner-model>-><worker-model>.
- Do I need separate credentials?
- No — Captain calls configured provider routes through the existing OpenAI-compatible LLM adapters; it does not implement OAuth or a second credential store.
Alternatives
lhmd/dsh-director-toolkit · lhmd/dsh-promotion-toolkit · ztl34245881-commits/dsh-task-planner