Plugins

Records the version of the edit target as an observation when the agent inspected the file with bash instead of the read tool, so the filesystem observation policy stops rejecting edits with FS_NOT_OBSERVED. Applies to edits only; blind whole-file overwrites stay guarded.

Converts Office and PDF files in the DSH workspace to Markdown locally — no network, no API quota — before the model reads them. Writes each .md next to its source so only a file path enters the context, and returns byte size, line count, token estimate and a heading outline, for a single file or a whole directory.

Local art pipeline for DSH: drives the locally installed SD WebUI and ComfyUI backends through ten image_* tools and an Art Studio sidebar panel with an embedded ComfyUI canvas, post-processes with the Python shipped with those installs (rembg cutout, Real-ESRGAN upscale, seamless tiling), and writes Godot-ready assets with res:// paths once a project root is set.

Inspect and tidy a real Android phone over adb: storage surveys, per-app data sizes read from dumpsys diskstats without root, cache clearing that preserves app data, directory measurements and screenshots — five tools; reads never modify anything, writes require explicit confirmation.

ZCode-style enhancement of dsh plan mode: a model-initiated enter_plan_mode tool, hard interception of write-class tools at dispatch time (ctx.tools.guard) while planning, and a four-phase planning prompt section (explore, design, review, exit), reusing the official plan/mode state machine and exit review flow.

White-box memory in two parts: four user-authored long-term tiers (must-follow rules, behavior settings, user profile, optional rules) injected as one system-prompt section, plus a separate temporary working set (cross-session handoffs, stage goals, waiting items) with its own 1200-character budget that stops being injected once an entry is done. The long-term tiers are plain Markdown files the user can read and edit; the agent can only propose them, and nothing is written until the user accepts in the Chinese panel. Temporary entries are written only when the user explicitly asks, and live in <DSH_HOME>/temp-memory.json (created on first write), not in the memory directory. No auto-capture, no background writes, no network.

Double-click one of your own messages to rewrite it and continue in a new branch from that point.

Adds a Forget action to every session row menu, confirmed by a second click in the menu, that removes the session from the sidebar and from disk: running work is stopped, workspace, archive and pin records are detached through official channels, the log directory and projection-cache entries are deleted, and a revive guard re-deletes anything a late write or a lingering handle recreates; memories, attachments and other sessions are left alone.

Connect OpenCode Zen free-tier models into DeepSeek Harness. Real-time model discovery, per-model toggle in settings, one-click provider write. No API key required.

Session budget gate for DeepSeek Harness: prices every turn against the official price table, warns at 80%, stops the task at 100%, and writes a handoff snapshot so the next session can continue the work.

A prompt optimizer in the composer input tool row: pick a category and a template, then ✨ rewrites the draft in place with one-click undo. Pick from the Basic, Context and Image categories; you can cancel a run, and it shows elapsed time and token usage. Ships 15 templates covering task-instruction rewriting, step planning, safety-research framing, system-prompt role cards, and text-to-image / image-to-image prompts; context templates read the recent turns of the session for coreference resolution. Calls go through the host model service, and a settings section exposes the model and the budgets. Output guards check that user-written constraints survive, warning rather than auto-retrying. Adapted for dsh 0.1.7 and 0.2.

Adds a button to the composer that rewrites the unsent draft into a clearer, more specific prompt, with one-click restore, a settings page and an optional per-request model override.

Git repository connector for DeepSeek Harness: saved repo profiles addressed by name, work-copy operations (clone/fetch/pull/push/commit), Gitea/Forgejo/GitHub API tools, keys in the DSH credential center, and per-repo safety flags (TLS bypass, overwrite, force push).

Self-organizing multi-agent clusters with no orchestrator. Real DSH subagents coordinate through a shared append-only log that each agent reads and writes on its own schedule, with no topology and one dispatch per agent; a snapshot mode is also included where the harness freezes each neighbour view into the prompt and drives a ring/mesh/swarm cluster to convergence.

Skill browser and switchboard for the DSH web GUI: a sidebar entry lists skills from configurable source directories as cards, clicking a card opens the full SKILL.md, and the per-card switch writes disable-model-invocation and registers or removes that skill in the agent skill catalog, so the model can load exactly the skills that are switched on. It also recognises WorkBuddy expert-team folders and shows avatar, tags, quick prompts and members, registering its lead prompt as a skill when switched on. Source directories are editable in the panel.

Read and write Apple iCloud Calendar over CalDAV through five tools: list calendars, read a window with recurring events expanded, create an event, and delete behind a dry run that must confirm the exact object count. Credentials come from DPAPI, Keychain or libsecret, or from an environment variable.

One-button prompt optimizer: rewrites the composer draft via a configurable strategy, with an editable template, few-shot examples, pinned models and reasoning effort.

Adversarial review for DeepSeek Harness: put one settled plan on the stand, send read-only red-team experts at it by dimension, then split their findings into single defects you endorse or reject one by one. It never rewrites the plan and never fixes code.

Rewrites the draft in the DSH composer into a clearer, more executable prompt through the harness's own LLM routes, showing an original/optimized comparison before you adopt it.

Adds three settings pages to DeepSeek Harness: add, edit, enable and remove MCP servers; write agent skills by hand or install them from a GitHub repository; and keep a library of global personas with one active.

Reads the session event stream, with no extra model calls, and turns corrections, failures and fixes into pending lesson candidates. Every candidate is judged against written criteria and carries the reason it passed or was refused; only confirmation writes a skill, and only through the single validated writer `learn_skill_manage`. An idle curator marks unused skills stale and archives them by moving rather than deleting, and spares any skill that is still being loaded. Five tools: `learn`, `learn_review`, `learn_curator`, `learn_skill_manage`, `learn_skills`. Registers a prompt section stating the criteria, and a tool guard that refuses direct edits to skill files so the writer cannot be bypassed.

Reasoning-effort gear presets for third-party models: 16 vendor/model presets with per-mode custom wire values, written into dsh-llm-pi-ai with pre-write validation and per-rule rejection; per-model switching between OpenAI Chat Completions, OpenAI Responses and Anthropic Messages with automatic route regrouping; plus a composer Optimize button and a settings section that rewrite the input text with a model you pick.

Turns a ZSpace (极空间) NAS into native agent tools: list, stat, search, read, download, upload, write, create folders, rename, move, copy and delete files on the NAS. Works across networks through the desktop client relay — no LAN, no WebDAV, no SSH — and switches to the faster direct WebDAV transport automatically when the NAS is on the same network. Deletes require an explicit confirm: true; write, upload, create-folder, rename, move and copy run directly, and zspace_write overwrites a same-named file unless overwrite is false.

Generative UI for DeepSeek Harness — the agent writes TSX, dsh web renders it live, inline in chat and in a canvas panel