AI Agent Hub
Back to plugins
⚙️

dsh-turnsnap

Workflow Updated 2026.08.28

Run the following command in DeepSeek Harness:

dsh plugin install yongshuai0314/dsh-turnsnap

Paste the following prompt into your AI chat to install this plugin:

Run dsh plugin install yongshuai0314/dsh-turnsnap inside DeepSeek Harness to install this plugin; full source is at https://github.com/yongshuai0314/dsh-turnsnap .

About this plugin

Long agent sessions in DeepSeek Harness can rewrite dozens of files across many turns. When a turn goes wrong you are either left digging through git log for the last known-good state or simply losing the trail. dsh-turnsnap automates the checkpoint: after every completed agent turn, if the working directory is a git repository with uncommitted changes, it stages everything and creates a single tagged [turnsnap] commit whose message includes the turn number, file count, and timestamp, giving you a rollback anchor you never had to request manually.

The plugin is zero-config with no client half to set up. Once loaded, it hot-reloads within seconds and starts guarding every session in a git workspace. Checkpoints are serialized on an internal queue so that a parent session and its subagents closing turns simultaneously cannot race the git index. Three argument-free tools - turnsnap_pause, turnsnap_resume, and turnsnap_status - let you tell the model to pause or resume auto-commits mid-session, or call them from any script that can invoke harness tools for fine-grained control.

It suits developers running multi-turn code generation, refactoring, or batch tasks in DSH who want a one-step way to roll back safely. Be aware that checkpoints run git add -A with --no-verify, so pre-commit hooks are skipped; if you keep a deliberate partial stage across a turn boundary, pause first. Ignored files remain ignored, and subagents sharing the same workspace land in the same commit.

Use Cases

  • Fast rollback to a specific turn after multi-turn code generation
  • Avoiding git index races when subagents mutate the same repo concurrently
  • Per-turn auditable snapshots during large-scale refactoring or rewrites

Best For

  • Developers running multi-turn sessions in DeepSeek Harness
  • Maintainers of experimental projects that need frequent rollback
  • Automation users who want fewer manual git steps between turns