shared-handoff-dsh

Usage

Once installed there is no command to remember — the skills are triggered conversationally and the model loads the right SKILL.md itself.

task-id-bootstrap: open a task

Just say in a dsh chat (adjacent Chinese/English punctuation both work):

新开task-id=init-kmp，然后开个 init-kmp 分支

The model runs the bundled bootstrap script; success is proven by three lines:

Task state: .../.agents/state/tasks/init-kmp
Current task: init-kmp
Session binding: /Users/you/.dsh/sessions/.../session.jsonl.zstd

Progress then lives in .agents/state/tasks/init-kmp/process.md; say 继续，task-id=init-kmp later to resume. A directory without a session binding is only a partial result — the model must report it as such.

If Python (3.9+) is missing on first use, the model reports the gap and shows the install command for your platform, installing only after your explicit consent — never silently.

handoff: export / resume a session

When a thread gets long and you want a fresh one, say:

帮我做个 handoff

You get a paste-ready next-thread prompt (workspace / branch / done / verification status / next step). In the fresh session, open with:

继续上次 handoff

The model rebuilds context from state files instead of chat history. Phrases like 交接, 新开线程继续, 继续上次, and resume trigger it too.

The two skills cooperate

When a task-id is active in the same repo, handoff treats .agents/state/tasks//process.md as the canonical state instead of inventing a parallel one.

Cross-agent handoff

The state layout is identical to the Codex and Claude editions: a handoff exported from dsh can be resumed in Codex or Claude and vice versa (all three share the same session-tasks.json).