플러그인

Auto-titles conversations after the first user+assistant exchange, with a chat header button to update or manually edit titles.

Agent-facing keyword search over local DSH session logs: session_search returns matching sessions with timestamps and excerpts, session_read dumps one conversation, and both skip injected context such as memory snapshots; no index is built.

Layered context management: at each turn end older tool output becomes a deterministic card, the model can fold older ranges into its own summaries in up to three tiers (compress), and every original is archived locally for decompress and search_context. Works alongside DSH-Mem0: superseded runtime-context snapshots (such as recalled memories) are never archived, carded or folded, and are replaced by a one-line placeholder at turn end when the prompt cache is already broken, when the saving outweighs the cache miss, or above 75% of the window; a strataBudget service (pressure, allowance, countTokens) and a strata/fold-landed event are provided for other context plugins; strata_status warns when DSH's automatic compaction is off or its official summary keeps failing. Its L1 layer, archive rotation, prune guard and preset takeover are forked from dsh-live-compact 1.0.6 (MIT); uninstall dsh-live-compact or billion-context first — they replace the same compaction rows and Strata will not take over while either is installed.

Compare two or three answers in one session and drop chosen turns from every later model request.

Lists this session's runtime errors — turn errors, tool failures, command failures and subagent failures — as copyable cards in a conversation tab.

Makes DSH's OpenAI Responses calls stateful: after the upstream confirms it stored a response, each turn sends only the newly added input items plus previous_response_id, and a stateless upstream is detected on the first response so every later request falls back to the full history.

Gives hand-declared models on custom or relay routes a vendor-matched reasoning-effort picker: detects Claude, GPT, DeepSeek, GLM, Kimi, Qwen, Gemini, Grok, Doubao or MiMo from the model id, display name or endpoint host, fills in only the levels that vendor accepts plus the pi-ai compat fields its request format needs, and writes them into the profile config so the Models page reasoning switch turns on. Models you edit or switch off are left alone, and a Models page card hands them back on request.

Runs one task through several configured models at once — in parallel or as a primary/secondary review — and returns each model's answer plus an aggregated digest.

Estimated CNY API cost for each conversation turn and for the session token total, using DeepSeek peak/off-peak and per-model pricing; subagent costs are listed separately.

One click beside the composer turns a half-formed draft into a structured prompt the AI can act on: goal, constraints, output format and acceptance criteria filled in, placeholders only where a real decision is missing. Rewritten in place and never sent — undo with one click or Ctrl+Z. Four editable presets (general / code / writing / translation), optional conversation context to resolve references, and a model that either follows the session or is pinned to a chosen provider and reasoning effort. Requests survive session switches and typing while a rewrite is in flight; a result that no longer matches the draft is kept for retry instead of overwriting it. Pure JS, no build, no bundled dependencies.

Intercepts DSH tool results before the model reads them: dedupe cache for repeated calls, a declarative decision graph that replaces large outputs with structured digests, an injection gate, and an honest token ledger that includes a negative-return account for over-compression. Measured 1400-1600 tokens saved per hit; the cumulative ledger is currently negative and the README says why, including that it cannot skip model turns.

Gives each Workspace a stable tint from a warm-to-cool ramp, sets its name as a section heading above indented Sessions, and turns a finished unread chat's row green.

A conversation view beside Chat and Trajectory that turns a DSH session into a character stage, with Codex Pet sprites, comic bubbles and a completion report listing the files and attachments the turn produced.

A-share and OTC fund dashboard forked from SuCriss/dsh-leekbox (MIT): indices, market boards, limit-up sentiment thermometer, dragon-tiger list, full-market search, a watchlist with in-place position editing and price alerts, a stock detail window with K-line, money flow and F10, a signal screener, 7x24 news, AI stock analysis, and an OTC fund tab with NAV, estimated NAV and a fund detail window showing stage returns, a NAV chart and the disclosed top-10 holdings with live quotes. It keeps upstream's /api/leekbox/* routes and $DSH_HOME/.leekbox-*.json data files, so install it instead of SuCriss/dsh-leekbox rather than alongside it.

Google Play store screenshot generator for DeepSeek Harness — /playstoreshots turns an app logo plus two reference sheets into four 1080x1920 screenshots and one 1024x500 feature graphic via Bailian wan2.7-image-pro.

DeepSeek Harness plugin: nested turn folders and Finder-style show/hide for chat turns, without creating a new Session.

Inject soulbridge context into DSH sessions and feed user turns back as observations.

Blocks tool calls that create temporary files or directories in the workspace root, moves new root-level temp entries into tmp/_quarantine at turn end, and provides a graphical settings page.

Keeps a model request alive through a dropped connection: transport failures retry until the network returns while real errors still fail fast, and a turn killed by a process crash offers a one-click, state-checking continuation.

Counts down to 23:30 on workdays in a strip under the composer of the DSH web UI, turning yellow under one hour and red under ten minutes, plays a three-chime alert at the deadline with snooze and dismiss buttons, and a mute toggle persisted in localStorage.

Blinks the Windows notification-area icon the way QQ and WeChat do for an unread message — the app's own icon alternating with a fully transparent frame — and flashes the taskbar button, when a turn finishes or a session waits for an approval or an answer.

Registers bundled full-stack, frontend-design and browser-automation skill documents as four read-only tools; a focus argument selects which reference file is returned alongside the skill body.

Runs Hindsight long-term memory on your own machine: after a one-time uvx --from hindsight-embed@latest hindsight-embed profile create server setup (~450 MB of Python dependencies) and with your own LLM key, the plugin manages the local memory server itself (health probe, start, stop, logs), registers ten hindsight_local_* tools for status, page search, page read, recall, remember and reflect, injects a knowledge-page index once per session, writes each finished turn back into the bank, and extracts facts with your own LLM key. Do not enable it alongside the official @vectorize-io/hindsight-coding-agents plugin: both read the same bank and would inject duplicate context. At session end it writes a handoff snapshot that the next session injects as previous progress. The data directory is configurable (dataDir) and scripts/relocate.mjs moves an existing install to another drive, creating directory junctions where the server hard-codes its paths.

Wall-clock session and turn timing for the DSH Web GUI: a timer chip in the session header shows the session total, every finished turn shows its own duration above the action row, and a sidebar panel totals today, the last 7 days, the last 30 days or all sessions. The numbers are folded from turn/start and turn/end in the session log, so paging, compaction and restarts do not change them.