Plugins

Hermetic composition lock: fingerprints the profile's dependencies, bundle order, lockfile and patch layer; booting a drifted composition warns, or fails under strict mode (the --frozen posture for CI).

Safety guard that makes self-killing commands impossible: a safe_kill tool that always excludes the caller and every ancestor process, a bash pre-execute check for pkill self-matches, sudo -S stdin grabs, rm -rf / and similar patterns, and an extensible rule library with block / warn / hint severities.

Emails you when a long turn finishes or errors, when a goal completes or blocks, and when an approval or a question has waited past a timeout; each new notification deletes the previous one from the inbox over IMAP, so the inbox holds at most one.

Removes the model-capability gate that blocks image prompts: patches llm.resolveModelInfo results and adapter model tables to declare image input, so routes with no declared input modality can send and read images. No app.asar or config-tree edits.

Pixel-level screenshot diff for visual regression. Two local PNGs in; a machine-readable first line (`PIXEL_CHANGED=true|false`, greppable as an assertion), then the difference ratio, per-region bounding boxes, maximum channel delta and mean absolute error, plus a self-contained HTML comparison page. An ignore-region list with absolute or fractional coordinates excludes clocks, cursors and other per-render noise. It deliberately does no content recognition: it answers only "did this change", which is the part that can be made deterministic. Entirely local — no model call, no network, no upload.

Gives the agent its own clock: per-session time readings whose precision the agent can change at runtime, elapsed-time counters, and one-shot countdowns that survive a host restart.

Inlines an image block as base64 only in the one request where it first appears, and marks that same block offloaded on every later request so DSH's own image-offload projection renders it as the placeholder text carrying the read-only local path. An occurrence is keyed by session id, message id, content-tree path and attachment id, and a request is only counted after it produces its first chunk, so retries do not count as sent. Requests carrying a purpose are neither rewritten nor counted, a later read_image of the same file is a new occurrence and inlines once more, and only that request's copy is rewritten, so nothing is written back to the session.

Deletes a session from the sidebar session menu after a confirmation dialog, removing its log directory, projection-cache record and workspace/pin/archive accounting; registry calls are timeout-guarded so a stuck workspace registry cannot block the removal.

ChatGPT-like conversation layout for the DeepSeek Harness Web client: a 768px centered column, right-aligned neutral bubbles for user turns, full-width assistant text at 16/28, and rounded code blocks with their own copy banner. Adds semantic inline markup (`==confirmed==`, `!!risk!!`, `++key word++`, with bare 4xx/5xx codes highlighted automatically) and line-start check/warn/tip callouts; a fast/thinking chip in the composer that switches the model reasoning effort through ctx.modelDirectories; and sidebar contrast, where the selected session gets a tinted row with an accent bar and workspace folder icons are colored.

Markdown input for the composer that keeps the typed Markdown as the sent text while drawing bullets and checkboxes, styling bold, italic, inline code, strikethrough, headings, quotes and fenced code blocks, continuing lists on Shift+Enter, and wrapping selections with Cmd/Ctrl+B, I and E.

Adds two visual pages to DSH Settings. MCP servers are kept as a JSON source of truth that rewrites the profile patch managed block, with YAML validation and timestamped backups, plus import of a pasted mcpServers JSON that first lists what will be added or overwritten. Thinking unlock probes each OpenAI-compatible gateway for the reasoning efforts it really declares and writes the per-level config back while leaving other hand-written YAML byte ranges untouched; the available model list syncs from a separate button that previews additions and removals, then writes levels and models in one atomic pass, and can also run on the page schedule.

Pushes DSH session events to phone IM and lets you reply straight under that message, injecting the text back into the right session. Covers nine event kinds: task completion, execution error, execution blocked, manual abort, token limit reached, abnormal interruption, waiting for your answer, permission request, and waiting for plan confirmation. Channels are QQ single-chat bots and Feishu custom apps, both over WebSocket long connections — no public IP, domain or tunneling needed. Several bots can be configured at once, each subscribing to the sessions it cares about. Approvals and questions arrive with tappable buttons, and replies route home through three layers (button, quoted reply, explicit short-id prefix), falling back to the most recent intervention-needed notification and echoing which session it went to.

Cross-session and cross-project long-term memory: a non-blocking silent sedimentation dispatch on turn/end writes into a local SQLite store (FTS5 trigram full-text index with insert/update/delete sync triggers) that later turns recall from; entries are scoped per project alongside a global tree, and the plugin ships its own embedded web dashboard (it starts the HTTP server itself, no separate process) for browsing, graphing and controlled deletion.

Copies a session log path (or a paste-ready handoff block) and turns the multi-frame zstd log into a readable file another session can continue from.

Registers a configurable prompt variable so an agent preset whose persona uses complete: true can still receive an injected text block; the plugin ships no content of its own and reads the text from its config.

22 light-only palettes (Pantone / Chinese traditional / Morandi) and 14 composable surface effects for DeepSeek Harness, with twelve-shichen clock rotation, manual picks that lead until the next boundary, workspace binding, per-session colour marks and a nickname/avatar row whose editor lives in the settings page.

Ports CSDN blog article typography into the DeepSeek Harness web GUI — heading hierarchy, quote blocks, solid-grid tables, inline code and Atom One syntax highlighting, in matching light and dark palettes.

Live token usage and estimated cost in the sidebar: the current task plus every session combined, an as-of clock for those totals, per-session time detail (elapsed, turns/steps, first-token, model vs tool time, decode speed), and a per-turn breakdown with each turn cost on hover.

Injects conclusions already established in the previous reasoning block after each tool result, so the model continues from them instead of re-deriving the same conclusions.

Plugin updater for DeepSeek Harness that delegates every version decision to the dsh CLI, so the pnpm release-age policy still decides what is eligible; it snapshots before applying, can roll back, keeps a lock list, and can check or auto-apply on a schedule the user turns on.

Convert Notion blocks and Markdown in both directions, with no API client and no external dependency. A Notion paragraph is an array of rich-text runs while Markdown says one string, so the two directions are a split and a merge rather than a pair of inverses — rich-text runs are cut at every annotation boundary one way and re-merged the other, with fences widened when the content already contains backticks. Reports every lossy step instead of swallowing it.

Ships a skill for verifying a change with the cheapest check that still yields a definite answer: an L0-L3 cost ladder from static checks to a single suite, condition polling with a timeout instead of fixed sleeps, assertions that can actually fail, and durations that are measured rather than remembered. The package also carries the two zero-dependency Node tools the skill calls - wait-until.mjs (poll a command, a file, or a port until it is ready) and measure-runtime.mjs (wall-clock timing for any command, plus a segmented timer for library use).

Prices tokens with the official model price table, keeps the per-session and global running cost beside the composer context meter, mirrors your DeepSeek account balance in the sidebar, blocks a turn once this period passes your own global ceiling (clamped to that balance) or a per-session ceiling (each session sets its own, independently), and resets the billing baseline per session or globally — the two are independent, so resetting one never resets the other — so the period goes back to zero while history only rolls into the lifetime total.

Delegates self-contained coding subtasks to the local Claude Code subscription through the official Agent SDK and returns the result to the host session. Nineteen per-call parameters control model, thinking effort, permission mode, allowed tools, turn count and budget cap; background jobs return a jobId immediately and stream incremental output; the result carries a session handle that can be passed back as resume to continue the same conversation; commands expected to run past ~20 seconds go through the built-in mcp__dsh__shell channel, which streams output line by line without consuming model context; a Claude Code monitor panel in the session lists jobs, renders the structured process, shows live output blocks and cancels a run; the delegation procedure ships as a runtime-registered skill, so installing the plugin installs the skill.