dsh-superagent

DeepSeek Harness (dsh) is an open-source agent harness developed by DeepSeek AI. It uses an architecture in which everything is a plugin and is powered by Cordis, whose design is described in “A Programming Paradigm for Spatiotemporal Composability.” The npm package can launch a local Web UI with `npx @deepseek-ai/dsh web`, serving at `http://127.0.0.1:3080` by default and opening the default browser; it can also be built and run from a repository checkout. The project is in developer preview and warns that compatibility-breaking changes may occur. It is licensed under the MIT License, with third-party dependencies and their licenses documented separately.

dsh-purge is an open-source cybersecurity red-team plugin for the official DeepSeek Harness installed locally. It patches matching files in the local @deepseek-ai packages, backs up originals, overrides the host shim, and injects an encrypted default prompt or a user-selected prompt-inject.md at session start; its ruleset logic can also select local AGENTS.md or CLAUDE.md files and apply a workspace baseline policy. A dock beside the session title provides Clean and Drill pages, grouped patch management, prompt and ruleset editing, skills management, authorized own-server scope controls, and Apply, Restore, Revert, or Uninstall operations. It supports the official DeepSeek Harness web profile and desktop executable separately, with the latest releases targeting dsh 0.2. The repository states that maintenance is paused, and describes the project as non-profit software distributed under the MIT license for controlled evaluation of a host the user is authorized to administer.

DSCODE is a macOS terminal coding agent and CLI built on DeepSeek Harness. Its persistent shell retains the working directory, environment, background jobs, edits, searches, patches, and test runs, while the TUI, CLI, and scripts share one session runtime. Sessions can hand tasks to other sessions, run side questions in read-only child sessions, delegate work to child agents in isolated Git worktrees, and use an independent read-only model to review changes and approve permissions. The `dscode exec` command supports non-interactive scripts and CI, with output streamed to stdout and tool activity and session identifiers sent to stderr. It also supports MCP servers, triggers, cross-session memory, multiple model providers, and a workspace-write sandbox with human-approval and automatic-review modes. Distribution is available through npm, Homebrew, GitHub release tarballs, and source; the project requires macOS 14 or later and is licensed under MIT. It is an independent community project, not an official DeepSeek product.

An experimental community preset for DeepSeek Harness, an LLM-agent harness. It starts a session with the real Minimal tool schema containing only a shell tool and a file-reading tool, using that condition to anchor the model's initial trajectory. After the first durable tool call or assistant reply, it promotes the session to the full Standard tool catalog, making heavier tools available on demand. The repository also includes related zero-anchored, self-introduction, prefab, and eternal-Minimal modes, and states that it is not an official DeepSeek preset or affiliated with DeepSeek.

DeepSeek and Destroy (DSD) is a long-horizon coding-agent skill for executing large implementation plans through persistent worker loops. A premium parent agent handles architecture, decomposition, decisions, and phase gates, while fresh lower-cost specialist workers implement, review, repair, and recheck repository-scale changes. The default worker backend is external OpenCode using opencode-go/deepseek-v4-flash, and the skill is designed for Codex, Claude Code, OpenCode, Kilo Code, and comparable coding harnesses. Each task is rendered into an immutable JSON contract and managed through attempt directories containing prompts, reports, logs, state, and scope baselines. Its normal loop uses an Implementer, a fresh independent Reviewer, and—when needed—a fresh Fixer followed by another Reviewer. Python-based gates check objective integrity such as hashes, lifecycle, and exact source movement; an optional read-only Evidence Clerk interprets already-produced evidence for the parent but cannot invent proof, rerun verification, repair code, waive integrity failures, approve work, or recurse into another Clerk. Documentation is split between always-relevant parent rules and cold-loaded workspace, harness, transport, compaction, prompt, and specialist-role materials.