akqwpeter-prog/dsh-agent-conductor
Dispatch self-contained tasks from DSH to 11 external agent CLIs (Codex, Claude Code, TraeCode, OpenCode, Gemini, Cursor, Kimi, Qwen, Copilot, WorkBuddy, Grok): a host-only bundle registering the conductor_dispatch tool plus an auto-triggered conductor skill.
Lets the DeepSeek Harness agent recognise when a job is better done by another coding agent and delegate it: by skill-description matching the model runs a ~90-line zero-dependency dispatch.py that starts a self-contained task in one of 11 external agent CLIs (Codex, Claude Code, TraeCode, OpenCode, Gemini, Cursor, Kimi, Qwen, Copilot, WorkBuddy, Grok) in headless mode, then brings the stdout result back into the conversation as the answer. Task text goes only to the target CLI's own provider; keys stay local. An optional bundle exposes the same registry as a first-class conductor_dispatch DSH tool.
Install
mkdir -p .dsh/skills/conductor && cp -R skills/conductor/. .dsh/skills/conductor/Skill-based install: copy skills/conductor/ into any skill root (project-level .dsh/skills/ or global ~/.dsh/skills/). No restart needed — the agent auto-recognises the skill by description matching from the next message. An optional conductor_dispatch tool bundle is also available for profile installation. Repo akqwpeter-prog/dsh-agent-conductor, verified 2026-08-31.
Compatibility
Any DeepSeek Harness profile; zero runtime dependencies (Python stdlib only for dispatch.py); dispatches to Codex, Claude Code, TraeCode, OpenCode, Gemini, Cursor, Kimi, Qwen, Copilot, WorkBuddy, Grok — each uses its own login quota; MIT.
Details
- Repo: akqwpeter-prog/dsh-agent-conductor
- Category: Tools & Capabilities
- Stars: 7
- Version: repo main (skill; no npm package)
- Last push: 2026-08-24
- First seen: 2026-08-15
Recent updates
Skill-based delegation to 11 external agent CLIs; auto-triggered by description matching; read-only dispatch script; optional conductor_dispatch tool bundle; zero runtime dependencies.
FAQ
- Do I need an API key for this?
- No — each target CLI uses its own existing login quota (Codex/Claude Code/OpenCode etc. authenticate through their own mechanisms). The plugin adds no extra key.
- How does the agent decide when to delegate?
- The SKILL.md description matching lets the model recognise delegation-worthy requests (e.g. "have Codex translate this") and run dispatch.py automatically — no manual trigger needed.
- Is my task text sent to a third party?
- The task text goes only to the CLI's own provider (the one you asked it to use); credentials stay local and are never forwarded.
Alternatives
aaravarr/dsh-subagent-max · shaokeyibb/dsh-plugin-product-subagents · zemy17-dsh-subagent-agy