@aiwayds/dsh-llmwiki-memory
dsh-plugin：面向 dsh 的 OKF v0.2 主题记忆——本地优先、由 Git 跟踪的捆绑包，热路径无 LLM 注入，带后台提炼的两阶段观察器
插件会安装到这里；不确定时保持 web。
npx -y @deepseek-ai/dsh plugin --profile web add @aiwayds/dsh-llmwiki-memory@0.3.0Llmwiki Memory 以 @aiwayds/dsh-llmwiki-memory 发布，当前版本为 0.3.0。Plugin Hub 会校验它的 manifest，并保存精确安装来源，便于复现安装结果。
English | 中文
A dsh plugin: maintains "working topic memory" as an OKF (Open Knowledge Format v0.2) knowledge bundle, persisted in a local git repository (optionally synced to a private GitHub repo), with conclusions traceable through git history, sessions automatically observed and distilled into knowledge, and relevant topics injected to the model before every turn.
Long sessions forget. Cross-session, even more so. This plugin maintains structured topic memory: each Topic records a matter's name, dependencies, open questions, current conclusion, impact, and recommendations. When a conclusion changes, edit the file and commit — git log directly answers "when, by whom, and why did this conclusion change".
markdown + YAML frontmatter concept document (type: Topic) that the whole OKF ecosystem (Obsidian, OKF validators) can consume directly; ships with the provenance (sources), trust (generated/verified), and lifecycle (status/stale_after) field families.topic_history tool and /wiki history make change history first-class.~/.dsh/llmwiki/), zero config, zero credentials; setting repo enables GitHub sync (single repo, single bundle, single main, write-through + debounced push; rebase conflicts are demoted and flagged for a human — no automatic smart-merge).depends graph walk), millisecond-scale; zero matches = zero injection; per-topic digest ≤300 tokens, top-K ≤4, total budget ≤1.5k tokens — all configurable./wiki stats reports hit rate, top-N, near-miss distribution, and tuning suggestions — tune from evidence, not vibes.depends (machine-readable directed edges) plus body [[wikilinks]] and markdown links (human-written edges) form one graph; retrieval walks it in both directions (per-level decay, configurable depth) so a single hit pulls in a knowledge subgraph; every write rebuilds the meta/backlinks.json reverse index, and /wiki show lists "who references me, and how" — check the blast radius before changing a conclusion.topic_observe; a background distill lane (session end + every N turns, model configurable) distills them into formal Topics in batches; when the model itself deems something worth keeping, it topic_saves directly./wiki onboard — native dsh ask-user panels walk you through the five decisions: mode / repo / distill model / injection tier / auto-observe — nothing is written until the final confirm;topic_save; /wiki status for health, /wiki stats for injection stats.| Model tools | Purpose |
|---|---|
topic_save | Distill/revise a Topic (name / dependencies / open questions / conclusion / impact / recommendations) |
topic_observe | Jot an atomic observation (decision/finding/constraint/question), pending distill |
topic_search | LLM-free keyword search over memory |
topic_history | A topic's conclusion change history (git log as a tool) |
| Command | Purpose |
|---|---|
/wiki onboard | Interactive setup wizard on dsh-native ask-user panels (mode / repo / distill model / injection tier / auto-observe); typed fallback where no ask-user UI exists |
/wiki status | Bundle health: topic count, observation backlog, conflicts, sync status |
/wiki stats | Injection stats: hit rate, top-N, near-miss distribution, tuning advice |
/wiki list / show / history | Browse topics, backlinks, and change history |
/wiki graph | Generate a relationship-graph web page (force-directed, draggable/zoomable, hover for conclusions) and open it in the browser |
/wiki sync [pull|push] | GitHub mode: manual pull/push (automatic by default) |
/wiki config / set <key> <value> | View and edit config (thresholds, budgets, distill model, …) |
dsh plugin --profile <your profile> add @aiwayds/dsh-llmwiki-memory
First thing after installing: run /wiki onboard. The bundle lives at ~/.dsh/llmwiki/ by default ($DSH_LLMWIKI_HOME overrides). GitHub sync: /wiki set repo <owner/name> (suggested repo name dsh-wiki-memory, to keep it distinct from the plugin's own source repo); credentials come from $GITHUB_TOKEN or a logged-in gh CLI (login is not this plugin's job).
First-time setup belongs to /wiki onboard; day-to-day tuning is /wiki set <key> <value> (writes the llmwiki namespace in settings.yaml, effective from the next session). All keys and defaults:
| Key | Default | Meaning |
|---|---|---|
repo | empty (local-only) | GitHub sync repo owner/name; suggested dsh-wiki-memory; empty = back to local-only |
autoInject | true | Per-turn injection master switch |
injectDedup | true | Session-level injection dedup: topics already injected in this session are not re-injected (registry cleared at session end; budget-dropped topics stay injectable; deduped topK slots are NOT backfilled) — ADR 0012 |
topK | 4 | Max topics injected per turn |
perTopicBudget | 300 | Per-topic digest token budget |
totalBudget | 1500 | Total injection budget per turn |
matchThreshold | 0.3 | Hit threshold; tune from /wiki stats near-miss evidence |
tagBoost | 0.15 | Additive boost per tag hit |
graphDepth | 2 | depends graph walk depth (0 disables) |
recencyWindowDays | 7 | Recency bonus window (+0.2) |
autoObserve | true | Capture atomic observations every turn |
includeSubagents | true | Whether injection and observation also engage subagent sessions (ADR 0011); off skips them entirely |
observationMaxChars | 2000 | Per-side per-turn observation truncation |
distillProvider / distillModel | empty (distill off) | Distill lane model route; both must be set to enable. With a UI, /wiki set distill-provider / distill-model without a value opens a picker panel (provider list → that provider's model catalog); a mixed provider model / value for splits into both keys |
This project's shape is directly inspired and supported by:
agent/inbox/spliced + systemPrompt.context()) follows the mechanism it validated on real dsh.include-subagents on), injection and observation apply to subagent sessions too. /wiki set include-subagents off skips delegated sessions entirely — no injection, no observation, no distill triggers; the topic tools stay on the global layer, so an explicit topic_save from a child still lands. Out-of-process subagents (claude-code/codex providers) never load this plugin anyway.meta/distill-state.json records each lane's outcome, checkable via /wiki status./wiki set and settings.yaml edits take effect most reliably from the next session start./wiki onboard splits the distill decision into two dependent questions (provider first, then that provider's model catalog), pre-validated with resolveModelInfo — a provider with no live route blocks and re-asks, an off-catalog model (a non-NO_ADAPTER failure: outside the advisory catalog, possibly still usable) warns but is allowed; hosts without an ask UI or a usable model route fall back to typed input. The same validation backs the /wiki set picker panels.MIT
provider/modeldistill-modeldistillEveryTurns | 20 | Distill every N turns of a long session |
distillOnSessionEnd | true | Distill once when a session ends |
pushDebounceSeconds | 45 | GitHub-mode debounced push interval |