Plugins

Compaction ledger for the DeepSeek Harness: audit every context compaction (what was shadowed, when, at what token cost) and let the model recall pruned content on demand

Native shodh-memory integration for DeepSeek Harness — persistent cognitive memory without an MCP server in the context window.

Hands secrets and PII to the agent without them entering its context, with checksum-validated extraction and masked content classification.

Chat quotes for DSH: select text in any message to quote or comment; the quote waits in a Chat quotes chip on the composer (never inside the draft), expands on hover into selected text and your comment, and folds into the message only when you send.

Power-Up — small, opt-in upgrades for DeepSeek Harness. Adds a right-click action menu to the Trajectory ledger, an optional stop reason in the session header's background jobs panel (appended to the injected completion notice), and an install control for

Per-conversation behavioral rule fence for DeepSeek Harness: rules live in one JSON file, are edited from a Settings page or the conversation panel, and every rule carries its own on/off switch so a single rule can be muted without deleting it (muted rules are skipped at injection and the remaining numbering stays contiguous); any number of conversations can also be excluded as a whole so they receive no fence at all, and picking conversations never touches the rules themselves; the enabled rules are injected as one runtime-context section at a fixed order so the system prompt prefix stays stable.

Audit how each turn's context budget is spent. Breaks down what a turn pays for AGENTS.md, the skill catalog, tool schemas and runtime context; flags repeated injection of unchanged content as pure waste; ranks skill-catalog bloat (with 60+ skills installed, how many tokens of descriptions does the model read every turn) and tool-schema inflation per plugin. Deterministic — no LLM, so a wrong answer is a rule bug with a name. Four tools including a cross-session sweep.

Audit what context compaction actually lost. Reads the raw session event log and checks session-constraint survival (the standing instructions the user gave — do-not-X, must-Y, only-in-Z — and whether the summary still carries them; arXiv 2608.11242 measured ~17% survival across compactors), summary hallucination (paths, versions, ports and identifiers the summary invents that were never in the history it replaced), ghost references ("see above" pointing at removed text), token accounting (a summary nearly as large as what it claims to free), and unaccounted prunes. Four agent tools including a cross-session sweep that ranks conversations by how much compaction hurt them. Deterministic — no LLM, so a wrong answer is a rule bug with a name.

IDEA-style Git workbench for the Web UI: commit graph with lanes and ref labels, log filters, commit details, split and unified diffs with per-hunk staging, a fine-grained zero-context mode, local changes with amend and sign-off, rename-following file history, per-line annotate, stash, branch operations, compare, tags, an interactive rebase planner and conflict resolution. Records which model produced a commit by stamping an Assisted-by trailer from the live session, and renders any existing commit trailers as chips, so AI-authored history stays traceable.

Read-only CNKI bibliographic search for DSH: four MCP tools (search, detail, export, status) drive your own Edge/Chrome through a native-messaging bridge, with no PDF downloads, a 3-second minimum interval and a 300-action daily cap. Requires pip install wensu plus wensu setup and loading the browser extension before the four tools appear.

Adds a Timestamp toggle to the composer; while it is on, each message you submit is followed by a hidden context message carrying the current local time, so the model knows when you sent it. The switch is remembered per conversation.

Adds Tako chat models to the official llm-pi-ai provider list from Tako's public catalog, with live refresh, context windows from size tags, and browser one-click sign-in writing a cr_ key into DSH credentials.

Bailian (DashScope / Model Studio) model catalog in the DSH settings page: lists every model the configured API key can reach, groups them by model family, probes whether a model is actually callable (an unopened model answers 403, which DSH reports as an invalid key), refuses to add models that fail the probe, and shows the active session's context window and remaining budget. An sk- key has no balance endpoint, so quota here means context quota.

Quantify what switching projects costs. Measures each workspace's cold-start overhead (AGENTS.md + skill catalog + tool schemas + runtime context), its steady state, and how many turns it takes to reach it — then compares workspaces and attributes the difference. Attribution requires at least three workspaces and says so instead of inventing a correlation from two points; a factor with no spread across workspaces is reported as a property of the profile, not of the project. Recommends which workspaces to keep warm, which to archive, by entry cost times visit frequency. Deterministic.

See what a conversation actually costs: how many model requests, how big the system prompt and tool table are, cumulative tokens, estimated spend, and how far you are from automatic context compaction. Reads the local session log only; never calls a model, never adds a word to the system prompt. Token and spend figures are estimates from character counts, not provider usage.

Adds a "Terms" action to finished assistant answers that explains the computer-science terms it detects with Chinese definitions, occurrence counts, highlighted context snippets, and clickable nested explanations from a built-in offline dictionary, with no model calls.

Context-based next-message suggestions and draft rewriting for DeepSeek Harness, with candidate paging, shortcut recording, and companion MCP tools.

Story Mode preset for DeepSeek Harness: role-play GM persona, three-way choices, .md character cards and world book, context distillation and prompt wrapping.

Percentage-only compaction and model-configurable GPT and DeepSeek token counting for DSH

Token-efficient web reading for DeepSeek Harness: one call fetches several pages through a real Chrome/Edge over CDP, gates them by relevance and returns a single digest, cutting context tokens by 80-98%. Real search URLs, login flow, screenshots, three proxy modes.

A possibly useful DSH token usage plugin: cross-session stats on where your tokens went - four headline figures, context length distribution, a 12-week heatmap and weekly trend, plus workspace and model rankings; it also ships a next-turn token forecast and a draft-optimization feature.

White kitten accountant on the window sill: shows the tokens and money for each answer and for the whole session, the money the prompt cache saved, the context size against the model window, the next peak/off-peak price change and the live DeepSeek balance; it purrs when stroked, reacts while the agent works and synthesizes all of its sounds.

Shortcut buttons in the composer row that send preset text into the current conversation in one click. Button label, size, opacity, shape and payload are editable in place, and the whole group moves left or right of the context meter by right-click dragging, persisted in localStorage. The group takes only the width it needs, so the context meter and the usage area stay clickable.

A run_tests tool that turns a failing test run into precise context instead of a raw dump. It detects the runner (pytest, jest, vitest, cargo, go, npm), extracts the file, line and assertion of each failure, and stops the loop when failures with an identical signature repeat. Output is decoded so Windows-encoded and non-Latin assertion text stays readable.