AI Agent Hub
Back to plugins
dsh-safe-workflow preview

dsh-safe-workflow

Workflow Updated 2026.08.19

Run the following command in DeepSeek Harness:

dsh plugin install cnwutianhao/dsh-safe-workflow

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

Build the project from https://github.com/cnwutianhao/dsh-safe-workflow first, then run dsh plugin install cnwutianhao/dsh-safe-workflow from the DSH source checkout to add it to the Web profile.

About this plugin

When an AI coding agent makes changes, the hardest part is often not what it writes but whether you can trace, verify, and roll back what it did. dsh-safe-workflow turns each coding task into an explicit contract: start it, checkpoint the workspace, verify with a recorded command, and close only when acceptance checks pass. Every mutating tool call is intercepted by a pre-execution approval gate, so nothing changes until you say so. It is not a sandbox but a policy-and-evidence layer that makes each agent step auditable and reversible. It is built for developers who want a structured, evidence-backed workflow on top of their existing DSH agent setup, without giving up the flexibility of a host-process plugin.

Screenshots

Use Cases

  • Snapshot the workspace before an agent batch edit, verify after, and restore if acceptance fails
  • Gate every mutating tool call in a shared repo so no critical path is touched without approval
  • Record verification command output as an evidence chain before committing agent-generated code

Best For

  • Developers using DeepSeek Harness agents who need auditable change trails
  • Tech leads in team repos who want approval gates on every mutating AI operation
  • Engineers seeking structured workflows and best-effort rollback without sacrificing agent flexibility