dsh-team-runner

The raw subagents API in DeepSeek Harness gives you low-level provider calls such as spawn, fork, and acp, but composing a multi-role workflow from scratch means hand-wiring context, signaling, and lifecycle management every time. dsh-team-runner collides that layer into three declarative commands: team_run executes a full team pipeline, team_list inspects configured teams, and team_delegate hands the current task to the next member—so a planner, writer, reviewer chain is just a teams entry in your profile.

On the policy side, hooks under tools/pre-execute let you gate every delegation with explicit checks and fail loud when a condition is not met, preventing silent fallbacks. A maxConcurrentDelegations cap keeps parallel fan-out under control, and the provider setting (spawn, fork, or acp) lets you match whatever agent runtime your Harness instance is configured for.

This plugin is aimed at developers and teams already running multi-agent workflows on DeepSeek Harness who want to stop re-rolling ctx.subagents.start() calls and instead declare roles and prompts once, letting the plugin handle orchestration, signal passing, and disposal.