AI Agent Hub
Back to plugins
🧩

dsh-tool-surface-proof

admin-security Updated 2026.08.26

Run the following command in DeepSeek Harness:

dsh plugin install dongsheng123132/dsh-tool-surface-proof

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

Run dsh plugin install dongsheng123132/dsh-tool-surface-proof in your DeepSeek Harness terminal to install this plugin; source code is available at https://github.com/dongsheng123132/dsh-tool-surface-proof

About this plugin

After an upgrade, can you prove that the tool set the model actually sees still matches the approved baseline? dsh-tool-surface-proof is built to answer exactly that question. It does not generate tool catalogs or compress MCP inventories; instead it verifies deployment-specific, explicitly recorded surface envelopes in a DeepSeek Harness deployment, checking every tool identity and schema against ordered nameHash, schemaHash, and position locks across agentScope, permissionMode, and presentationMode.

The verifier detects nine drift categories: missing or unexpected surfaces, scope, permission, or presentation mismatches, added or removed tools, schema changes, optional order drift, stale or future-dated observations, duplicate identities, and revision mismatches. Reports carry only SHA-256 digests, counts, and drift labels — never tool names, descriptions, raw schemas, or business content — making them safe to retain as long-lived, tamper-resistant audit evidence.

The CLI reads a single workspace-relative regular file, rejects symlinks and path traversal, performs no network requests or subprocesses, and writes one content-addressed JSON report under an explicit artifact directory using exclusive creation with read-back verification. Both the DSH plugin bundle and a standalone stdio MCP server expose the same interface-free core. If you run production DeepSeek Harness deployments and need concrete before-and-after proof at every upgrade, this plugin is a focused, low-surface-area fit.

Use Cases

  • After a DeepSeek Harness upgrade, verify that the model-visible tool set still matches the approved baseline
  • Generate content-addressed SHA-256 reports as long-lived audit evidence that tool surfaces have not changed
  • Before shipping a new deployment revision, confirm every surface lock still holds across agent, permission, and presentation scopes

Best For

  • Platform engineers operating DeepSeek Harness deployments in production
  • Security and compliance teams that need audit trails for tool-surface changes
  • Ops teams managing multi-scope (agent, permission, presentation) tool deployments