DeepSeek Harness Plugins · admin-security
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
Offline verification that independently operated builders produce byte-identical specified artifacts from identical declared inputs, using hash-only receipts with no build execution or network access.
Verifies that the recorded model-visible tool surface of a DeepSeek Harness deployment still matches an approved lock across agent, permission, and presentation scopes, producing content-addressed audit reports.
A security-scoped DeepSeek Harness plugin adding precise code read/write, atomic range edits, AST structural search, and read-only LSP queries without modifying the host.
A static plugin to control which tool schemas are injected into the DSH model context, featuring a settings page and a tool_visibility manager to save context space.
User-configurable per-model pricing with time-band multipliers, live cost projection across all sessions, and a 24-hour multiplier strip in the composer dock.
Adds a Codex section to DeepSeek Harness Web settings, enabling ChatGPT login via the official authorization service and the openai-codex model route.
Offline deterministic verification that a recorded build's hash-only access receipt remained within its declared file, environment, network, clock, and randomness closure.
Enables HTTPS for remote browser access via Let's Encrypt IP certificates without a domain, injects polyfills to unlock the settings page and privileged RPCs, with automatic renewal and login redirect.
A dsh-TUI plugin that locally blocks high-confidence credentials (API keys, private keys, tokens) before model delivery, storing no prompt content and reporting categories only.
A shared settings-registration platform for DSH plugins: eight control types, three scopes, namespaced isolation, and one unified settings page grouped per plugin.