dsh-agent-lines
Community plugin · Apache-2.0

Let one AI lead
run a team for you

You state the goal. The lead splits it into research, architecture, UI, development, test, acceptance, release, and ops lines; assigns work, tracks progress, and reports back. Close the window or restart DSH; the work and messages remain.

curl -fsSL https://raw.githubusercontent.com/Lencen/dsh-agent-lines/main/install.sh -o install.sh && bash install.sh
First use: create a manager line and 8 member lines, then receive acks
demo profile · generic labels · no internal data

What it takes off your plate

No repeated context

Each member line is a long-lived DSH Session with its own context. Send new work to the right role instead of re-explaining the project.

No manual dispatch

The lead decomposes work, assigns it, coordinates, resolves disagreements, and collects results. You read the summary.

No lost work

Tasks and messages are stored locally. Restart DSH and continue; member state and unread messages are still there.

Built for long-running projects and multi-role work. For a one-off subtask, DSH's built-in subagent / Agent Teams is simpler.

A real usage pattern

1. You state the goal

“Add CSV export to the reports page, with tests and a short runbook.”

2. The lead forms the team

Split into research, architecture, UI/UX, development, test, acceptance, release, and ops; create each line.

3. The team works in parallel

Members do their parts; the lead tracks progress, resolves interface disagreements, and reassigns when needed.

4. You get the result

The lead reports conclusions, file paths, risks, and open questions. You do not chase every member yourself.

First use: let the lead form the team

You are the development team lead: own decomposition, assignment, coordination, tracking, and summary.
First bind this Session with line_claim_manager, then use line_create for:
research / architecture / uiux / dev / test / acceptance / release / ops
After creation, confirm with line_list, then wait for my first task.

Restart DSH Web after install or update. Technical details, protocol, and limits: README / architecture. Community plugin, not an official DeepSeek product.