AI Agent Hub
Back to plugins
⚙️

dsh-proof

Workflow Updated 2026.08.25

Run the following command in DeepSeek Harness:

dsh plugin install EvilIrving/dsh-proof

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

Run dsh plugin install EvilIrving/dsh-proof in your DeepSeek Harness terminal to install the plugin; the full source is at https://github.com/EvilIrving/dsh-proof .

About this plugin

When a DeepSeek Harness agent finishes its last tool call, the turn simply commits with no objective check that the work is truly done. dsh-proof closes that gap: before each top-level turn closes, it spawns a read-only verifier subagent, collects a structured verdict, and injects any gaps and evidence back into the driving agent so it can address them.

The verifier is locked to read-only tools (read, grep, glob), has a zero-recursive-depth cap, and can never mutate files or spawn subagents. Passing turns add zero tokens; a failing turn appends exactly one bounded message plus a short follow-up line, keeping the KV cache append-only. If the verifier service is unavailable the plugin degrades to a silent no-op, never blocking the user's turn.

Best for developers who rely on the Harness for multi-step coding, debugging, or research workflows and want an independent, auditable acceptance gate instead of trusting the agent's self-declared completion.

Use Cases

  • Prevent premature turn closure in multi-step coding tasks
  • Add independent evidence checks to research workflows
  • Objectively review fixes before accepting a debugging session

Best For

  • Developers running multi-step tasks on DeepSeek Harness
  • Teams that need automated acceptance gates without manual oversight
  • Platform engineers who want an auditable trail over agent output