DeepSeek Harness Plugins · Workflow
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
Let agents write self-contained HTML artifacts that render in a side tab, stored outside the workspace and persisting across projects.
A deterministic frontend verification loop that lets the agent open, read, assert layout facts, interact, screenshot, and re-verify its own build before reporting work as done.
Adds Windows desktop automation to DSH: screenshot, local OCR, mouse/keyboard control, and Paint drawing with zero third-party dependencies.
Scans Makefile, package.json, justfile, and deno commands declared in a project and registers them as ready-to-run agent tools.
An acceptance-driven auto-completion engine for DeepSeek Harness that independently verifies tasks via deterministic checkers before certifying completion, with repair loops and regression detection.
A disk-persisted, state-machine-driven execution planning plugin with template generation, dependency tracking, audit logging, and debrief notes that survive session restarts.
Parses OpenAPI 3.x specs to generate typed TypeScript or Python clients and detects breaking changes between versions, fully local with no external dependencies.
A deterministic, zero-LLM context triage plugin for DeepSeek Harness that automatically prunes stale, repeated, failed, and oversized tool outputs to save token budget in long sessions.
Upgrades the DSH sidebar to a Project/Task dual mode where tasks run in standalone directories without workspace binding, managed via panel or /task command.
A safer, evidence-backed coding workflow plugin for DeepSeek Harness with task contracts, workspace checkpoints, and pre-execution approval gates.
Registers a preset_run host-plane tool that spawns an isolated agent session with a given preset, executes a task, and returns the final answer — a headless equivalent of the Web new-session flow.
A zero-dependency, offline document skill pack providing README generation, PR description, changelog curation, and code review as pure-prompt agent skills.