99 results
Per-conversation x-opencode-session header injection for OpenCode Go in DeepSeek Harness (DSH). Fixes 400 MissingSessionID, keeps prompt-cache routing optimal, scoped strictly to opencode.ai, runtime-switchable without restart.
Keep long answers and detailed prompts side by side with a dockable DSH Composer.
A DeepSeek Harness plugin that actively assesses artifact value and keeps agent workspaces organized, auditable, and recoverable.
DSH bundle that bridges the Codebase Memory MCP code knowledge graph into DSH and can keep indexed repositories fresh with an optional debounced filesystem watcher.
Agent Skill for DeepSeek Harness (dsh): list.
Long-task gatekeeper for DeepSeek Harness: detects one-shot mammoth requests, asks whether to plan first, and scaffolds plan.md + a capped goal before the agent starts
Recursive Self-Improvement (RSI) engine for DeepSeek Harness: a bounded perceive → integrate → verify → repair → propose loop that keeps a downstream desktop app in sync with the fast-moving upstream harness. AI for AI — DSH maintains DSH.
Let the model render an image into the DSH web chat flow (QQ/WeChat style) via a show_image tool: the message content keeps only the path text, the image never enters model context, and the browser shows the picture inline, scrollable with history.
DSH port of pi-tokensaver: keeps the tokensave semantic graph fresh and teaches the agent to prefer its semantic tools
Opt-in detached watchdog for the DSH Web process with optional snapshot-checked repair and explicit patch restoration.
Codex-style /btw for DeepSeek Harness: ask a side question answered by a conversation-seeded subagent, keeping the answer out of the main context
/fuck — read your Claude Code & Codex conversation history in parallel, analyze your preferences and tool habits, migrate them into DSH as a persisted profile, and keep remembering them.
Fold DSH Web execution steps into compact expandable process boxes while keeping dialogue and final answers outside.
DSH plugin that auto-retries failed API requests — when third-party APIs return errors (rate limiting, service unavailable, etc.), intercepts and retries to keep the conversation going. Supports auto/manual mode with configurable status codes and error codes.
Keep DSH Web available and repair broken optional plugins
Keep Think reasoning rows visible and auto-expanded, and always force Chinese thinking and replies. 保持 Think 思考行可见并自动展开，并始终强制中文思考与回复。
Curate tool output before it enters the model's context: keep error/warning lines and head/tail, spill the full text to disk with a pointer. Faster turns, smaller context, sharper attention — for every tool, not just bash.
Run Claude Code, Codex, Pi, or the built-in loop side by side in one dsh — pick the engine per session, switch without restarting, keep one session store, history, and trace.
Compatibility antidote for dsh-claude-ux: keep its risk-control UI visibly enabled while neutralizing the actual debuffs.
Vision for text-only DeepSeek Harness agents: delegate image reading to a one-shot subagent on a configurable vision route (MiniMax / Kimi / any OpenAI-compatible provider), keeping image bytes and the vision model's context out of the main session.
Minimal-anchored agent preset for DeepSeek Harness: first request keeps the real Minimal tool pair, promotion unlocks the full Standard catalog with a one-shot background-jobs notice
A DeepSeek Harness (dsh) security plugin: blocks agents from reading or writing sensitive files (.env, credentials, key material), masks leaked secret-shaped values in tool results, keeps an audit journal, and exposes safe inspection tools that never print raw values.
ego lite for DeepSeek Harness, with a memory it keeps: the agent recalls what it learned about a site before it acts, promotes a script that worked into a reusable site tool, and hands you the keyboard through a real dsh prompt when a page needs a human.
DeepSeek Harness plugin: aligns markdown tables in assistant replies right after each turn (surface-replace, replay-safe) and adds a /clear-history command that empties the model-visible conversation while keeping the append-only log.