Skip to content
Back to search

llm-wiki

LLM-compiled knowledge base plugin with Project Knowledge Checkpoints Export, personal specialist skills, trusted adapters, Idea-to-Project workflows, compact queries, inventory, datasets, audits, sessions, research, and outputs.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/nvk--llm-wiki/badge.svg)](/listings/nvk--llm-wiki)

HTML

<a href="/listings/nvk--llm-wiki"><img src="/api/listings/nvk--llm-wiki/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

**Claude Code** (native plugin):

claude plugin marketplace add nvk/llm-wiki
claude plugin install wiki@llm-wiki

**OpenAI Codex** (marketplace plugin):

Install from GitHub:

codex plugin marketplace add nvk/llm-wiki
codex plugin add wiki@llm-wiki
# Start a new Codex thread, then use @wiki or type $ to select wiki-query

Install from a local checkout with the managed bootstrap helper:

./scripts/bootstrap-codex-plugin.sh --scope user --verify

Or register the local checkout manually:

codex plugin marketplace add /absolute/path/to/llm-wiki
codex plugin add wiki@llm-wiki

Canonical explicit invocation:

$wiki-query "What does the wiki say about hardware wallet threat models?"
@wiki research "hardware wallet threat models"
@wiki collect "bitcoin memes" --wiki memes-bitcoin
@wiki ingest https://example.com/article
@wiki audit --project coldcard-threat-model
@wiki session status
@wiki feedback list --unpromoted
@wiki session disable   # optional opt-out
@wiki ll "codex plugin install gotchas"

Upgrade:

codex plugin marketplace upgrade llm-wiki
codex plugin add wiki@llm-wiki

Remove:

codex plugin remove wiki@llm-wiki
codex plugin marketplace remove llm-wiki

Troubleshooting: - `codex plugin marketplace add` registers the catalog; `codex plugin add wiki@llm-wiki` installs and enables the cached plugin non-interactively. - Open `/hooks` to review and trust the bundled hooks if you want automated session capture. The `@wiki` skill works without hook trust. - `$wiki-query` is the small, explicit, read-only skill for lookups. In Codex CLI/IDE, type `$` or open `/skills` and select it. It never activates implicitly or changes wiki files. - `@wiki` is the full research and maintenance entry point. Natural-language wiki requests can still auto-activate it. - Restart Codex after changing config if an existing session does not pick up the new plugin state. - If you run Codex under a sandbox wrapper like `nono`, see [Nono Sandbox Permissions](#nono-sandbox-permissions) — Codex needs r+w to `$HOME/.codex` for plugin install. - On Windows, the plugin's bundled `commandWindows` hooks use a `.cmd` launcher and accept PowerShell UTF-16 JSON. No separate hook script setup is required.

**DeepSeek Harness** (native filesystem skill):

From a local checkout:

./scripts/install-deepseek-harness.sh --verify
# Start DeepSeek Harness and invoke /wiki

The installer copies a self-contained skill to `${DSH_HOME:-$HOME/.dsh}/skills/wiki`, the documented user skill root. Harness support intentionally uses the narrow filesystem-skill contract rather than a custom runtime plugin while DeepSeek Harness remains in developer preview. Session hooks are optional and can use the harness's official Codex-hook bridge separately.

**OpenCode** (instruction file):

Add to your `opencode.json` (project-level or `~/.config/opencode/.opencode.json` for global):

{
  "instructions": ["https://raw.githubusercontent.com/nvk/llm-wiki/master/plugins/llm-wiki-opencode/skills/wiki-manager/SKILL.md"],
  "permission": {
    "external_directory": {
      "~/.config/llm-wiki/**": "allow",
      "~/Library/Mobile Documents/com~apple~CloudDocs/wiki/**": "allow"
    }
  }
}

OpenCode fetches the URL fresh on every session start — no manual updates needed. If you prefer a local copy instead:

curl -sL https://raw.githubusercontent.com/nvk/llm-wiki/master/plugins/llm-wiki-opencode/skills/wiki-manager/SKILL.md > ~/.config/opencode/AGENTS.md

For a smaller read-only setup, use the best-effort query preset instead:

{
  "instructions": ["https://raw.githubusercontent.com/nvk/llm-wiki/master/plugins/llm-wiki-opencode/skills/wiki-query/SKILL.md"]
}

The OpenCode profile is sync- and budget-tested, but not tied to one model, so it does not have a provider-specific live quality gate. Treat it as a portable

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

.claude-plugin/marketplace.json · Manifest · 21 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

This file has changed since it was scanned, so this is not exactly the text that was scanned, and no line is marked. The findings are listed with the safety scan above.

  1. {
  2. "$schema": "https://anthropic.com/claude-code/marketplace.schema.json",
  3. "name": "llm-wiki",
  4. "version": "1.0.0",
  5. "description": "LLM-compiled knowledge base plugin with Project Knowledge Checkpoints Export, personal specialist skills, trusted adapters, Idea-to-Project workflows, compact queries, inventory, datasets, audits, sessions, research, and outputs.",
  6. "owner": {
  7. "name": "nvk"
  8. },
  9. "plugins": [
  10. {
  11. "name": "wiki",
  12. "description": "LLM-compiled knowledge base with Project Knowledge Checkpoints Export, personal specialist skills, trusted adapters, Idea shaping and Project promotion, compact queries, ingestion, inventory, datasets, audits, sessions, and outputs.",
  13. "version": "1.0.0",
  14. "author": {
  15. "name": "nvk"
  16. },
  17. "source": "./claude-plugin",
  18. "category": "productivity"
  19. }
  20. ]
  21. }
README.md · README · 924 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

This file has changed since it was scanned, so this is not exactly the text that was scanned, and no line is marked. The findings are listed with the safety scan above.

  1. ```
  2. ██╗ ██╗ ███╗ ███╗ ██╗ ██╗██╗██╗ ██╗██╗
  3. ██║ ██║ ████╗ ████║ ██║ ██║██║██║ ██╔╝██║
  4. ██║ ██║ ██╔████╔██║ ██║ █╗ ██║██║█████╔╝ ██║
  5. ██║ ██║ ██║╚██╔╝██║ ██║███╗██║██║██╔═██╗ ██║
  6. ███████╗███████╗██║ ╚═╝ ██║ ╚███╔███╔╝██║██║ ██╗██║
  7. ╚══════╝╚══════╝╚═╝ ╚═╝ ╚══╝╚══╝ ╚═╝╚═╝ ╚═╝╚═╝
  8. ```
  9. [llm-wiki.net](https://llm-wiki.net/) · [@LLMWIKI on X](https://x.com/LLMWIKI) · [github.com/nvk/llm-wiki](https://github.com/nvk/llm-wiki)
  10. - BTC bc1p9l6n4m6kyql0wxm6nu46ekw9wue2c85hk8yuw5z5487g2uzr0xcsqn8k75
  11. - ETH 0x103564a1c1fc3774D34050e5397869387e4cbEF6
  12. - SOL 2hRuXsz9gS5PrmdU6kxrFCz95ecMtTRZcX3abiow3udY
  13. LLM-compiled knowledge bases for any AI agent. Capture rough Ideas, research and shape them, then explicitly promote approved briefs into delivery Projects. Also includes parallel research, collector catalogs, session memory, source ingestion, compilation, audits, querying, and artifact generation. Ships for Claude Code, OpenAI Codex, DeepSeek Harness, OpenCode, and portable agents. Obsidian-compatible.
  14. ---
  15. [Install](#install) · [Quick Start](#quick-start) · [Sessions](#session-memory) · [Commands](#commands) · [How It Works](#how-it-works) · [Research Modes](#research-modes) · [Thesis Research](#thesis-driven-research) · [Query Depths](#query-depths) · [Linking](#linking-works-everywhere) · [Obsidian](#obsidian-integration) · [Architecture](#claude-first-multi-runtime) · [Nono Sandbox](#nono-sandbox-permissions) · [Upgrade](#upgrade) · [Changelog](#changelog) · [Credits](#credits)
  16. ---
  17. ## Changelog
  18. **v1.0.0** — **Stable identity, honest provenance, and five-runtime packaging.**
  19. Adds stable `wiki_id` registry checks and engine declarations, actionable raw
  20. coverage ownership, footnote/wiki-link integrity, safe nested raw namespaces,
  21. immutable refresh revisions with human-only verification, pinned Git
  22. manifest-only ingestion, native DeepSeek Harness skills, and Windows Codex
  23. hooks. The major version marks these provenance and identity contracts as the
  24. supported baseline without making existing optional fields mandatory.
  25. **v0.25.0** — **Portable multi-runtime reliability.** Adds configurable
  26. `llm-wiki lint --fail-on critical|warning|suggestion` exit gates, portable
  27. Claude/Codex/OpenCode packaging without `rsync`, safer Windows session-state
  28. and path handling, `HOME`-aware hub discovery, and isolated Codex installs that
  29. respect the selected `CODEX_HOME`.
  30. **v0.24.0** — **Project Knowledge Checkpoints Export.** Exports comprehensive,
  31. cross-topic project handoffs under `docs/knowledge/<slug>/` with
  32. dry-run-first create and refresh, read-only verification and bounded import,
  33. exact source and section coverage, explicit omissions, semantic privacy
  34. minimization, deterministic sealing, attested overrides, and a thin-output
  35. guard. Checkpoint writes never authorize commit, publication, or import.
  36. **v0.23.0** — **Personal specialist framework.** Adds optional user-owned,
  37. instruction-only `SKILL.md` review methods under the user's local wiki hub,
  38. explicit per-topic allowlists, deterministic validation and management, and research
  39. selection with version/hash provenance. The public release contains the
  40. framework only; personal specialist packages and wiki-derived candidate reports
  41. are not bundled or published.
  42. **v0.22.0** — **Declarative private-adapter routing.** Registered adapters
  43. may declare provider-neutral intent and exact-URL routes plus an adapter-owned
  44. workflow guide. The public plugin discovers the route before ingestion but no
  45. longer embeds any provider's authentication, browser, recovery, or editing
  46. workflow.
  47. **v0.21.3** — **Adapter-boundary transition.** Consolidates the v0.21 line and
  48. keeps the session-hook compatibility fix. Provider-specific workflow material
  49. from that line is superseded in current source by manifest-driven routing and
  50. lives only in the corresponding private adapter.
  51. ## Install
  52. **Claude Code** (native plugin):
  53. ```bash
  54. claude plugin marketplace add nvk/llm-wiki
  55. claude plugin install wiki@llm-wiki
  56. ```
  57. **OpenAI Codex** (marketplace plugin):
  58. Install from GitHub:
  59. ```bash
  60. codex plugin marketplace add nvk/llm-wiki
  61. codex plugin add wiki@llm-wiki
  62. # Start a new Codex thread, then use @wiki or type $ to select wiki-query
  63. ```
  64. Install from a local checkout with the managed bootstrap helper:
  65. ```bash
  66. ./scripts/bootstrap-codex-plugin.sh --scope user --verify
  67. ```
  68. Or register the local checkout manually:
  69. ```bash
  70. codex plugin marketplace add /absolute/path/to/llm-wiki
  71. codex plugin add wiki@llm-wiki
  72. ```
  73. Canonical explicit invocation:
  74. ```text
  75. $wiki-query "What does the wiki say about hardware wallet threat models?"
  76. @wiki research "hardware wallet threat models"
  77. @wiki collect "bitcoin memes" --wiki memes-bitcoin
  78. @wiki ingest https://example.com/article
  79. @wiki audit --project coldcard-threat-model
  80. @wiki session status
  81. @wiki feedback list --unpromoted
  82. @wiki session disable # optional opt-out
  83. @wiki ll "codex plugin install gotchas"
  84. ```
  85. Upgrade:
  86. ```bash
  87. codex plugin marketplace upgrade llm-wiki
  88. codex plugin add wiki@llm-wiki
  89. ```
  90. Remove:
  91. ```bash
  92. codex plugin remove wiki@llm-wiki
  93. codex plugin marketplace remove llm-wiki
  94. ```
  95. Troubleshooting:
  96. - `codex plugin marketplace add` registers the catalog; `codex plugin add wiki@llm-wiki` installs and enables the cached plugin non-interactively.
  97. - Open `/hooks` to review and trust the bundled hooks if you want automated session capture. The `@wiki` skill works without hook trust.
  98. - `$wiki-query` is the small, explicit, read-only skill for lookups. In Codex CLI/IDE, type `$` or open `/skills` and select it. It never activates implicitly or changes wiki files.
  99. - `@wiki` is the full research and maintenance entry point. Natural-language wiki requests can still auto-activate it.
  100. - Restart Codex after changing config if an existing session does not pick up the new plugin state.
  101. - If you run Codex under a sandbox wrapper like `nono`, see [Nono Sandbox Permissions](#nono-sandbox-permissions) — Codex needs r+w to `$HOME/.codex` for plugin install.
  102. - On Windows, the plugin's bundled `commandWindows` hooks use a `.cmd` launcher
  103. and accept PowerShell UTF-16 JSON. No separate hook script setup is required.
  104. **DeepSeek Harness** (native filesystem skill):
  105. From a local checkout:
  106. ```bash
  107. ./scripts/install-deepseek-harness.sh --verify
  108. # Start DeepSeek Harness and invoke /wiki
  109. ```
  110. The installer copies a self-contained skill to
  111. `${DSH_HOME:-$HOME/.dsh}/skills/wiki`, the documented user skill root. Harness
  112. support intentionally uses the narrow filesystem-skill contract rather than a
  113. custom runtime plugin while DeepSeek Harness remains in developer preview.
  114. Session hooks are optional and can use the harness's official Codex-hook bridge
  115. separately.
  116. **OpenCode** (instruction file):
  117. Add to your `opencode.json` (project-level or `~/.config/opencode/.opencode.json` for global):
  118. ```json
  119. {
  120. "instructions": ["https://raw.githubusercontent.com/nvk/llm-wiki/master/plugins/llm-wiki-opencode/skills/wiki-manager/SKILL.md"],
  121. "permission": {
  122. "external_directory": {
  123. "~/.config/llm-wiki/**": "allow",
  124. "~/Library/Mobile Documents/com~apple~CloudDocs/wiki/**": "allow"
  125. }
  126. }
  127. }
  128. ```
  129. OpenCode fetches the URL fresh on every session start — no manual updates needed. If you prefer a local copy instead:
  130. ```bash
  131. curl -sL https://raw.githubusercontent.com/nvk/llm-wiki/master/plugins/llm-wiki-opencode/skills/wiki-manager/SKILL.md > ~/.config/opencode/AGENTS.md
  132. ```
  133. For a smaller read-only setup, use the best-effort query preset instead:
  134. ```json
  135. {
  136. "instructions": ["https://raw.githubusercontent.com/nvk/llm-wiki/master/plugins/llm-wiki-opencode/skills/wiki-query/SKILL.md"]
  137. }
  138. ```
  139. The OpenCode profile is sync- and budget-tested, but not tied to one model, so
  140. it does not have a provider-specific live quality gate. Treat it as a portable
  141. best-effort preset and keep OpenCode's write and shell permissions disabled for
  142. query-only sessions.
  143. The `external_directory` permission is required because the wiki hub lives outside the project directory. Set the paths to match your hub location. Alternatively, use `--local` mode (`.wiki/` in the project) to skip permissions entirely.
  144. Web search requires `export OPENCODE_ENABLE_EXA=1`.
  145. **Pi** (skill file, best for local models):
  146. Pi's minimal system prompt leaves room for on-demand wiki workflows on local
  147. models. Load the full skill for research and write-capable maintenance:
  148. ```bash
  149. pi --skill path/to/llm-wiki/plugins/llm-wiki-opencode/skills/wiki-manager/SKILL.md
  150. ```
  151. Invoke it as `/skill:wiki-manager`, or let Pi load it when the request clearly
  152. matches its description.
  153. For fast read-only queries with Pi's currently configured provider, use the
  154. generic launcher. It disables discovery and write tools and loads the compact
  155. shared query protocol:
  156. ```bash
  157. ./scripts/pi-wiki-query
  158. ```
  159. For DS4, the provider-specific launcher additionally creates an isolated Pi
  160. state directory and pins the local model settings:
  161. ```bash
  162. ./scripts/pi-ds4-wiki-query
  163. ```
  164. Set `PI_CLI`, `DS4_BASE_URL`, or `LLM_WIKI_PI_DS4_STATE_DIR` only when your
  165. local setup differs from the defaults. Both launchers accept `--dry-run` to
  166. show the exact command. The equivalent generic Pi settings are:
  167. ```bash
  168. pi \
  169. --append-system-prompt path/to/llm-wiki/profiles/query-lite/SKILL.md \
  170. --tools read,grep,find,ls \
  171. --no-extensions --no-skills --no-prompt-templates --no-themes
  172. ```
  173. The DS4 query profile is intentionally unable to write. Switch to the full
  174. skill for ingest, research, compile, lint, or other mutating workflows. See
  175. [`profiles/ds4/README.md`](profiles/ds4/README.md) and the reproducible
  176. [`benchmarks/README.md`](benchmarks/README.md) DS4 lane.
  177. **Any LLM Agent** (portable instruction file):
  178. ```bash
  179. # Read-only queries: small default
  180. cp profiles/query-lite/SKILL.md ~/your-project/AGENTS.md
  181. # Research and maintenance: complete protocol
  182. cp AGENTS.md ~/your-project/AGENTS.md
  183. ```
  184. The query-lite profile works with agents that can read and search files. The
  185. root `AGENTS.md` contains the complete write-capable protocol for agents that
  186. can also edit files and search the web.
  187. ## Claude-First, Multi-Runtime
  188. Claude Code is the principal user. Keep one shared behavior layer and thin packaging layers per runtime:
  189. - `claude-plugin/` is the primary distribution target and UX surface.
  190. - `claude-plugin/skills/wiki-manager/` is the behavioral source of truth.
  191. - `plugins/llm-wiki/skills/wiki/` is the generated Codex packaging target behind `@wiki`.
  192. - `claude-plugin/skills/wiki-manager/references/query-lite.md` is the canonical read-only query protocol.
  193. - `profiles/query-lite/SKILL.md` and generated `wiki-query` skills expose that protocol without the full research context.
  194. - `plugins/llm-wiki-opencode/` is the OpenCode and Pi packaging target.
  195. - `plugins/llm-wiki-dsh/` is the generated DeepSeek Harness filesystem-skill target.
  196. - `.agents/plugins/marketplace.json` makes the Codex plugin installable from this repo.
  197. - `AGENTS.md` is the portable single-file protocol for any other LLM agent.
  198. **Query and research presets:**
  199. | Client | Read-only query path | Full research path | Validation |
  200. |--------|----------------------|--------------------|------------|
  201. | Claude Code | `/wiki:query` command plus shared query protocol | Native plugin skill and commands | Deterministic fixtures plus live Sonnet/Opus runs |
  202. | Codex | Explicit-only `$wiki-query` (~2.8 KB) | `@wiki` (~14.5 KB before lazy references) | Deterministic fixtures plus app-server live runs |
  203. | Pi / DS4 | `scripts/pi-wiki-query`, or the isolated DS4 launcher, with read-only tools (~2.8 KB profile) | `pi --skill .../wiki-manager/SKILL.md` | Deterministic fixtures plus exact DS4 provider-payload measurement |
  204. | OpenCode | `wiki-query/SKILL.md` (~2.8 KB) | `wiki-manager/SKILL.md` (~30.0 KB) | Static budgets and generated-package sync; live model is best effort |
  205. | DeepSeek Harness | Load `/wiki` and request a read-only query | `/wiki` from `~/.dsh/skills/wiki` | Generated-package sync plus isolated install verification |
  206. | Any agent | Copy `profiles/query-lite/SKILL.md` (~2.8 KB) | Copy root `AGENTS.md` (~62.2 KB) | Static budgets |
  207. Sizes are checked-in UTF-8 bytes, not provider token estimates. Use query mode
  208. for lookups, evidence checks, lists, and inventory status. Use the full profile
  209. for research, ingestion, compilation, lint fixes, or any workflow that writes.
  210. Runtime mirrors are generated, not hand-maintained. Rebuild from the Claude source of truth:
  211. ```bash
  212. ./scripts/sync-codex-plugin.sh # regenerates plugins/llm-wiki/
  213. ./scripts/sync-opencode-plugin.sh # regenerates plugins/llm-wiki-opencode/
  214. ./scripts/sync-dsh-plugin.sh # regenerates plugins/llm-wiki-dsh/
  215. ./scripts/sync-query-lite-profile.sh # regenerates profiles/query-lite/
  216. ```
  217. Each sync script:
  218. - copies `claude-plugin/skills/wiki-manager/SKILL.md` into the target tree and reapplies a small list of runtime-specific wording patches
  219. - copies `references/` from the Claude source — references are runtime-neutral and shared verbatim (previously a symlink, now a real copy so Codex marketplace caching works)
  220. - generates compact `wiki-query` packages from the canonical query-lite reference
  221. - (Codex only) recreates `agents/openai.yaml` for Codex UI metadata and syncs the plugin version
  222. Drift is caught by `./tests/test-codex-sync.sh`,
  223. `./tests/test-opencode-sync.sh`, and `./tests/test-dsh-sync.sh`, which run the
  224. sync scripts and fail if generated packages differ from the source of truth.
  225. Practical rule: design workflows first for Claude commands and behavior, but keep the underlying knowledge model and references runtime-neutral. Runtime wrappers adapt invocation and metadata, not wiki logic.
  226. ## Nono Sandbox Permissions
  227. If you run any AI coding agent inside a [nono](https://github.com/always-further/nono) sandbox, the wiki needs filesystem access beyond the default profile.
  228. ### Claude Code / OpenCode
  229. ```json
  230. {
  231. "extends": "claude-code",
  232. "policy": {
  233. "add_allow_read": [
  234. "$HOME/.config/llm-wiki"
  235. ],
  236. "add_allow_readwrite": [
  237. "$HOME/Library/Mobile Documents/com~apple~CloudDocs/wiki"
  238. ]
  239. }
  240. }
  241. ```
  242. Replace `"extends": "claude-code"` with `"opencode"` for OpenCode.
  243. ### Codex
  244. Codex needs r+w to its own `$HOME/.codex` directory for plugin install, marketplace cache, state, and skill registration:
  245. ```json
  246. {
  247. "extends": "codex",
  248. "policy": {
  249. "add_allow_read": [
  250. "$HOME/.config/llm-wiki"
  251. ],
  252. "add_allow_readwrite": [
  253. "$HOME/.codex",
  254. "$HOME/Library/Mobile Documents/com~apple~CloudDocs/wiki"
  255. ]
  256. }
  257. }
  258. ```
  259. ### Path reference
  260. | Path | Access | Purpose |
  261. |------|--------|---------|
  262. | `$HOME/.config/llm-wiki` | read | Hub path config — checked first during resolution (v0.4.2+) |
  263. | Wiki data dir | readwrite | The wiki itself — use the actual path, not `$HOME/wiki` (see below) |
  264. | `$HOME/.codex` | readwrite | Codex only — plugin cache, skills, state, marketplace temp files |
  265. ### Hub resolution and `~/wiki`
  266. Hub resolution checks `~/.config/llm-wiki/config.json` first and only falls back to `~/wiki` when no config exists. If your wiki lives on iCloud or any non-default path, set the config and you don't need `$HOME/wiki` in the sandbox at all:
  267. ```bash
  268. # Set once — agents will resolve from config, never touch ~/wiki
  269. /wiki config hub-path "~/Library/Mobile Documents/com~apple~CloudDocs/wiki"
  270. ```
  271. Use a portable `hub_path` with `~` for iCloud-shared hubs. Older configs may
  272. contain `resolved_path`; llm-wiki treats it as a fallback cache because an
  273. absolute `/Users/<name>/...` path from one Mac will not be valid on another.
  274. The shared `wikis.json` should likewise store hub topic paths as
  275. `topics/<slug>`, not absolute user-home paths.
  276. If you prefer `~/wiki` as a symlink to iCloud, nono's Seatbelt follows symlinks — the target path must be allowed, not the symlink itself.
  277. ### Diagnosing access issues
  278. Without the right permissions, Seatbelt or macOS privacy controls can block file
  279. access. A useful diagnostic pattern is: `stat` succeeds for the iCloud wiki path,
  280. but reading `wikis.json` or listing `topics/` fails with `Operation not
  281. permitted`. That means the configured `hub_path` is correct; grant Full Disk
  282. Access or iCloud Drive access to the exact app launching the agent, restart it,
  283. and do not switch to a machine-local `resolved_path` or `~/wiki` fallback. Use
  284. `nono why` to diagnose sandbox rules:
  285. ```bash
  286. nono why --path ~/.config/llm-wiki --op read
  287. nono why --path ~/Library/Mobile\ Documents/com~apple~CloudDocs/wiki --op readwrite
  288. ```
  289. **OpenCode** also needs the `external_directory` permission in `opencode.json` (see [Install](#install)) — nono and OpenCode have independent sandboxes that both need the same paths allowed.
  290. ## Upgrade
  291. Agents and sandboxed sessions should use GitHub CLI web login with HTTPS git
  292. transport, not SSH. This avoids SSH host-key prompts and `known_hosts` writes
  293. inside nono:
  294. ```bash
  295. gh auth login --web --git-protocol https
  296. gh auth setup-git
  297. ```
  298. **Claude Code** — if `claude plugin update` pulls the latest correctly:
  299. ```bash
  300. git -C ~/.claude/plugins/marketplaces/llm-wiki remote set-url origin https://github.com/nvk/llm-wiki.git
  301. claude plugin update wiki@llm-wiki
  302. # Restart Claude Code to apply
  303. ```
  304. If the update command doesn't pick up the new version (stale marketplace cache), sync manually from the repo:
  305. ```bash
  306. # Clone or pull the latest
  307. git clone https://github.com/nvk/llm-wiki.git # or: git -C ~/llm-wiki pull
  308. # Sync plugin files to Claude Code's plugin cache
  309. REPO=~/llm-wiki/claude-plugin
  310. DEST=~/.claude/plugins/cache/llm-wiki/wiki
  311. VERSION=$(grep '"version"' "$REPO/.claude-plugin/plugin.json" | grep -o '[0-9.]*')
  312. rm -rf "$DEST"/*
  313. mkdir -p "$DEST/$VERSION"
  314. cp -R "$REPO/.claude-plugin" "$REPO/commands" "$REPO/skills" "$DEST/$VERSION/"
  315. # Restart Claude Code to apply
  316. ```
  317. **Codex** — upgrade from the marketplace:
  318. ```bash
  319. codex plugin marketplace upgrade llm-wiki
  320. codex plugin add wiki@llm-wiki
  321. ```
  322. **OpenCode** — if using the GitHub URL in `instructions`, updates are automatic (fetched every session). If using a local copy:
  323. ```bash
  324. curl -sL https://raw.githubusercontent.com/nvk/llm-wiki/master/plugins/llm-wiki-opencode/skills/wiki-manager/SKILL.md > ~/.config/opencode/AGENTS.md
  325. ```
  326. **DeepSeek Harness** — from an updated checkout:
  327. ```bash
  328. ./scripts/install-deepseek-harness.sh --verify
  329. ```
  330. **AGENTS.md** — just pull the latest and replace:
  331. ```bash
  332. curl -sL https://raw.githubusercontent.com/nvk/llm-wiki/master/AGENTS.md > ~/your-project/AGENTS.md
  333. ```
  334. Check your installed version:
  335. - Claude Code: look for the version in `/wiki` status output or check `~/.claude/plugins/installed_plugins.json`
  336. - Codex: run `./scripts/verify-codex-plugin.sh --scope user` and confirm the installed cache path and marketplace source match this repo
  337. > **New to a topic? One command, from anywhere:**
  338. > ```
  339. > /wiki:research "gut microbiome" --new-topic --min-time 1h
  340. > ```
  341. > Creates a topic wiki, launches parallel agents, and keeps researching for an hour — drilling into subtopics each round finds. Come back to a fully compiled wiki.
  342. ## Quick Start
  343. ```
  344. /wiki:research "nutrition" --new-topic # Create wiki + research in one shot
  345. /wiki:research "gut-brain axis" --wiki nutrition # Add more research to existing wiki
  346. /wiki:research "fasting" --deep --min-time 2h # 8 agents, keep going for 2 hours
  347. /wiki:research "keto" --retardmax # 10 agents, max speed, ingest everything
  348. /wiki:specialist suggest --wiki all # Rank reusable personal expert methods from current topic indexes
  349. /wiki:research "causal evidence" --specialist research-methodologist # Apply one enabled bounded review method
  350. /wiki:research "What makes long form articles go viral?" --new-topic # Question → decompose → playbook
  351. /wiki:thesis "fiber reduces neuroinflammation via SCFAs" # Thesis-driven: evidence for + against → verdict
  352. /wiki:thesis "cold exposure upregulates BDNF" --min-time 1h # Deep thesis investigation
  353. /wiki:collect "bitcoin memes" --wiki memes-bitcoin # Find, dedupe, download media, catalog, and optionally inventory artifacts
  354. /wiki:collect "bitcoin memes" --scale medium --media reference --inventory corpus # Catalog media without binary downloads
  355. /wiki:query "How does fiber affect mood?" # Ask the wiki
  356. /wiki:query "compare keto and mediterranean" --deep # Deep cross-referenced answer
  357. /wiki:query --resume # Where did I leave off?
  358. /wiki add https://example.com/article # Fuzzy router detects URL → ingest
  359. /wiki what do we know about CRISPR? # Fuzzy router detects question → query
  360. /wiki:ingest https://example.com/article # Manually ingest a source
  361. /wiki:ingest https://example.com/news --type articles --raw-subdir news/daily # Keep a typed source in a nested namespace
  362. /wiki:ingest --inbox # Process files dropped in inbox/
  363. /wiki:ingest-collection https://github.com/bitcoin/bips --wiki bitcoin # Bulk import spec repos
  364. /wiki:ingest-collection /path/to/repo --adapter git --manifest-only --revision v2.0 # Pin paths/blob SHAs without child copies
  365. /wiki:ingest-collection https://dump.bitcoin.it/dump_20260429_en.xml.bz2 --wiki bitcoin # Import MediaWiki dumps
  366. /wiki:ingest-collection messages.csv --adapter csv-messages --wiki bitcoin # Split message archives
  367. /wiki:ingest-collection "https://example.com/*" --adapter wayback-cdx --from 20100101 --to 20200101 # Import archived snapshots
  368. /wiki:inventory add ingest-candidate "Bitcointalk archive" --wiki bitcoin # Track source queues and next actions
  369. /wiki:inventory add item "TRX-4M ring and pinion" --wiki trx4m-1-18 # Track actual parts, tools, hosts, or assets
  370. /wiki:inventory list --view actions --limit 10 # Compact chat table of current inventory next actions
  371. /wiki:inventory scan-outputs --dry-run # Preview queues/backlogs before any inventory pivot
  372. /wiki:dataset add "Bitcointalk Temporal Graph" --location https://figshare.com/articles/dataset/BitcoinTemporalGraph/26305093 --wiki bitcoin # Index data that stays external
  373. /wiki:dataset list --view schema --limit 10 # Compact chat table of dataset schema/readiness state
  374. /wiki:dataset scan-outputs --dry-run # Find legacy data reports that could become dataset manifests
  375. /wiki:archive topic old-interest --reason "No longer active" # Preserve a topic but hide it from normal context
  376. /wiki:archive list --archived # Show active and archived topic wikis
  377. /wiki:archive restore old-interest # Bring an archived topic back
  378. /wiki:compile # Compile any unprocessed sources
  379. /wiki:audit --project gut-brain-playbook # Truth-seeking audit across outputs + wiki + fresh research
  380. /wiki:output report --topic gut-brain # Generate a report
  381. /wiki:output slides --retardmax # Ship a rough slide deck NOW
  382. /wiki:checkpoint create my-project --repo . --audience team # Preview a comprehensive cross-topic, privacy-sealed project handoff
  383. /wiki:checkpoint create my-project --repo . --audience team --apply # Stage, seal, verify, then write; never commit/push
  384. /wiki:checkpoint verify docs/knowledge/my-project # Recheck hashes and the privacy attestation
  385. /wiki:assess /path/to/my-app --wiki nutrition # Gap analysis: repo vs wiki vs market
  386. /wiki:lint --fix # Clean up inconsistencies
  387. scripts/llm-wiki identity status --wiki nutrition # Check stable config/registry identity
  388. ```
  389. When working from this repository checkout, you can also run deterministic
  390. checks without an agent:
  391. ```bash
  392. ./scripts/llm-wiki lint /path/to/wiki
  393. ./scripts/llm-wiki lint --fix /path/to/wiki
  394. ./scripts/llm-wiki schema status /path/to/wiki
  395. ./scripts/llm-wiki schema adopt /path/to/wiki # add starter topic guide
  396. ./scripts/llm-wiki schema adopt --dry-run /path/to/wiki # preview without writing
  397. ./scripts/llm-wiki archive --hub /path/to/hub topic old-interest --reason "No longer active"
  398. ./scripts/llm-wiki archive --hub /path/to/hub list --archived
  399. ./scripts/llm-wiki archive --hub /path/to/hub restore old-interest
  400. ./scripts/llm-wiki-session --hub /path/to/hub status
  401. ./scripts/llm-wiki-session --hub /path/to/hub disable # optional opt-out
  402. ./scripts/llm-wiki-session --hub /path/to/hub enable --mode balanced --tool-events 50
  403. ./scripts/llm-wiki-session --hub /path/to/hub rehydrate --cwd "$PWD"
  404. ./scripts/llm-wiki-session --hub /path/to/hub feedback list --unpromoted
  405. ./scripts/llm-wiki adapter add /private/adapter --read-root /private/input --write-root /private/results
  406. ./scripts/llm-wiki adapter doctor <id>
  407. ./scripts/llm-wiki adapter run <id> --request /absolute/request.json --json
  408. ./scripts/llm-wiki adapter run <id> --request /absolute/apply.json \
  409. --response /private/results/receipt.json \
  410. --approve-remote-write <plan-sha256> --json
  411. ./scripts/llm-wiki specialist init
  412. ./scripts/llm-wiki specialist create research-methodologist \
  413. --description "Reviews study design, causal claims, evidence quality, and reproducibility."
  414. ./scripts/llm-wiki specialist validate
  415. ./scripts/llm-wiki specialist enable research-methodologist --wiki nutrition
  416. ./scripts/llm-wiki specialist list --wiki nutrition --json
  417. ./scripts/llm-wiki checkpoint seal docs/knowledge/my-project --audience team --json
  418. ./scripts/llm-wiki checkpoint verify docs/knowledge/my-project --json
  419. ```
  420. This local helper covers structural checks and safe migrations that do not
  421. require an LLM. Checkpoint sealing is mandatory for generated project
  422. handoffs: it enforces comprehensive input-to-section coverage and a
  423. deterministic thin-output floor, hashes the bundle, scans for sensitive
  424. information without printing matched values, and writes `privacy-report.json`.
  425. There is no no-scan option; intentional exceptions name exact finding IDs and
  426. remain visibly attested as `overridden`. The agentic `/wiki:lint` workflow remains the full protocol for
  427. editorial and deep verification passes. The local schema helper creates only a
  428. starter advisory topic guide (`schema.md`); run librarian conventions advice
  429. for established wikis before adopting topic-specific vocabulary. The local archive helper
  430. performs the deterministic folder move plus `wikis.json`, hub index, and log
  431. updates.
  432. ## Personal Specialist Skills
  433. An optional user-owned library under `HUB/.skills/` stores reusable
  434. Agent Skills-compatible `SKILL.md` methods. Specialists encode bounded evidence
  435. and review protocols—not simulated CFO, physician, MBA, or PhD credentials.
  436. They are explicitly enabled per active topic, instruction-only in v1, and
  437. cannot grant tools, writes, subagents, or professional authority.
  438. Research can select zero to three enabled specialists (normally one), give each
  439. the same bounded evidence packet, and record its version and content hash in
  440. provenance. `/wiki:specialist suggest` examines current topic indexes and ranks
  441. candidate methods without creating them; `/wiki:specialist apply` performs a
  442. bounded review. See the [specialist reference](claude-plugin/skills/wiki-manager/references/specialists.md)
  443. for local storage, sharing boundaries, validation, selection, high-stakes
  444. escalation, and eval rules.
  445. ## Session Memory
  446. Long agent runs can lose useful context when a chat compacts, a terminal closes,
  447. or work moves between runtimes. LLM Wiki keeps a separate operational-memory
  448. layer under `HUB/.sessions/` (or `.wiki/.sessions/` for a local wiki) so a later
  449. turn can recover the thread without copying full transcripts into the knowledge
  450. base.
  451. | Feature | How it works |
  452. |---------|--------------|
  453. | **Capture** | Trusted hooks use balanced mode by default. They record harness metadata, current directory and Git context, small redacted events, per-session state, and Markdown digests. Full transcript bodies are not stored by default. |
  454. | **Session ID** | An ID such as `codex:abc123` combines the harness name with that runtime's native session ID. It is a local lookup key—not a secret, transcript, wiki, or topic ID. |
  455. | **Rehydrate** | `/wiki:session rehydrate` returns a compact context block and pointers to matching digests instead of bulk-pasting chat history. |
  456. | **Feedback** | High-signal corrections, preferences, approvals, and plan acceptance become reviewable candidates under `.sessions/feedback/`; generic acknowledgements are ignored. |
  457. | **Promote** | Digests and feedback remain operational memory until an explicit `promote` writes a distilled note to a topic's `raw/notes/`. Only then can normal compilation turn it into topic knowledge. |
  458. | **Opt out** | `/wiki:session disable` makes trusted capture hooks no-ops; `/wiki:session enable` turns capture back on. |
  459. ```text
  460. /wiki:session status
  461. /wiki:session list --limit 10
  462. /wiki:session show codex:abc123
  463. /wiki:session rehydrate --cwd "$PWD"
  464. /wiki:session promote codex:abc123 --topic meta-llm-wiki
  465. /wiki:feedback list --unpromoted
  466. /wiki:feedback promote fb-abc123 --topic meta-llm-wiki
  467. ```
  468. This harness-session layer is distinct from `.research-session.json` and
  469. `.thesis-session.json` crash recovery, and from `.session-events.jsonl` and
  470. `.session-checkpoint.json` workflow provenance. See the
  471. [session reference](claude-plugin/skills/wiki-manager/references/sessions.md)
  472. for the storage layout, privacy defaults, capture modes, and adapter contract.
  473. ## Commands
  474. | Command | Description |
  475. |---------|-------------|
  476. | `/wiki <natural language>` | Fuzzy intent router — say what you want and it routes to the right subcommand |
  477. | `/wiki` | Show wiki status, stats, and list all topic wikis |
  478. | `/wiki init <name>` | Create a topic wiki at `~/wiki/topics/<name>/` |
  479. | `/wiki init <name> --local` | Create a project-local wiki at `.wiki/` |
  480. | `/wiki:ingest <source>` | Ingest a URL, file path, PDF, or quoted text |
  481. | `/wiki:ingest <source> --raw-subdir <path>` | Keep a source in a nested namespace below its declared top-level raw type |
  482. | `/wiki:ingest --inbox` | Process all files in the topic wiki's inbox/ |
  483. | `/wiki:ingest-collection <source>` | Bulk-ingest Git doc repos, BIP-style proposal sets, MediaWiki dumps/API sites, message archives, or Wayback CDX snapshots |
  484. | `/wiki:ingest-collection <source> --adapter git\|mediawiki-dump\|mediawiki-api\|csv-messages\|wayback-cdx` | Force a collection adapter |
  485. | `/wiki:ingest-collection <source> --limit <N> --dry-run` | Preview or cap a large collection import |
  486. | `/wiki:ingest-collection <git-source> --manifest-only --revision <ref>` | Pin filtered Git paths and blob SHAs in one immutable manifest without copying child bodies |
  487. | `/wiki:adapter add <local-path>` | Register an existing explicitly trusted private adapter checkout; never clones or publishes it |
  488. | `/wiki:adapter list\|show\|doctor` | Inspect registrations and verify manifest/executable handshakes |
  489. | `/wiki:adapter run <id> --request <json>` | Execute a scoped v1 request and verify artifacts without automatic wiki import; remote writes additionally require an exact approved plan hash and private verified receipt |
  490. | `/wiki:adapter remove <id> --yes` | Remove only the machine-local registration, not adapter code or data |
  491. | `/wiki:specialist init\|create\|refresh\|list\|show\|validate` | Manage personal, instruction-only specialist `SKILL.md` methods under the local hub at `HUB/.skills/` |
  492. | `/wiki:specialist enable\|disable <name> --wiki <topic>` | Change the explicit per-active-topic specialist allowlist |
  493. | `/wiki:specialist suggest [--wiki <topic\|all>]` | Rank bounded specialist methods justified by current wiki indexes without creating them |
  494. | `/wiki:specialist apply <name> <question> --wiki <topic>` | Apply one enabled, validated specialist and report its version/hash |
  495. | `/wiki:collect "<things>"` | Find, dedupe, and catalog artifacts, examples, resources, media, memes, tools, entities, or source candidates |
  496. | `/wiki:collect "<things>" --scale tiny\|small\|medium\|large\|huge` | Control write behavior by operational scale, not just row count |
  497. | `/wiki:collect "<things>" --media archive\|thumbnail\|reference` | Download/cache bounded originals by default; use thumbnail for previews or reference to opt out |
  498. | `/wiki:collect "<things>" --inventory records` | Create per-item inventory records when the collected set is small enough to stay useful |
  499. | `/wiki:collect "<things>" --inventory corpus` | Track a large, unstable, or media-heavy collection as one corpus record linked to the catalog output |
  500. | `/wiki:inventory list` | List durable tracking records as compact chat-friendly tables or bullets |
  501. | `/wiki:inventory list --view actions` | Show current inventory next actions without dumping full records |
  502. | `/wiki:inventory add <kind> "title"` | Add an inventory record after checking that inventory is the right layer |
  503. | `/wiki:inventory save-view "name"` | Save a derived reusable table/list under `inventory/views/` |
  504. | `/wiki:inventory scan-outputs --dry-run` | Find old queue/backlog outputs and preview sample records before migration |
  505. | `/wiki:inventory migrate-output <path> --apply` | Additively create inventory records from a legacy output; never moves or deletes the output |
  506. | `/wiki:idea new [<slug>] "<seed>"` | Capture a rough proposal under `inventory/ideas/` without creating a Project |
  507. | `/wiki:idea list\|show` | Browse the topic or hub-wide Idea catalog with derived maturity and next actions |
  508. | `/wiki:idea develop\|shape <slug>` | Research an Idea, challenge assumptions, and prepare minimal/ideal shapes for approval |
  509. | `/wiki:idea promote <slug>` | Explicitly promote an approved snapshot into a linked Project with `WHY.md` and frozen `BRIEF.md` |
  510. | `/wiki:idea archive <slug>` | Archive the Idea record without implicitly archiving its linked Project |
  511. | `/wiki:portfolio [filters]` | Show a read-only hub-wide view of canonical Ideas and active Projects without creating a catch-all topic |
  512. | `/wiki:dataset list` | List dataset manifests as compact chat-friendly tables or bullets |
  513. | `/wiki:dataset list --view schema` | Show schema/readiness state without opening samples or data |
  514. | `/wiki:dataset add "title" --location <path-or-url>` | Add a dataset manifest without copying data into the wiki |
  515. | `/wiki:dataset profile <slug> --dry-run` | Preview lightweight profiling of size, format, headers, or schema observations |
  516. | `/wiki:dataset migrate-output <path> --apply` | Additively create dataset manifests from a legacy output; never moves or copies the underlying data |
  517. | `/wiki:archive list [--archived]` | List active topic wikis and optionally archived topic wikis |
  518. | `/wiki:archive topic <slug> --reason "why"` | Move a topic wiki to `topics/.archive/<slug>` and hide it from default context |
  519. | `/wiki:archive restore <slug>` | Restore an archived topic wiki to active status |
  520. | `/wiki:archive peek <query>` | Search archived topic indexes without reading archived articles |
  521. | `/wiki:session status` | Show automated session-capture mode, root, and indexed session count |
  522. | `/wiki:session disable` | Opt out of automated redacted session capture |
  523. | `/wiki:session enable --mode balanced` | Re-enable or tune automated capture and soft rehydration |
  524. | `/wiki:session capture` | Force a manual session digest checkpoint |
  525. | `/wiki:session rehydrate` | Print compact context from matching session digests |
  526. | `/wiki:session promote <id> --topic <slug>` | Promote a distilled digest into a topic raw note |
  527. | `/wiki:feedback list --unpromoted` | Review high-signal user-feedback candidates captured from session hooks |
  528. | `/wiki:feedback show <id>` | Inspect a candidate with distilled lesson, confidence, scope, and redacted preview |
  529. | `/wiki:feedback promote <id> --topic <slug>` | Promote a selected feedback candidate into topic `raw/notes/` |
  530. | `/wiki:compile` | Compile new sources into wiki articles |
  531. | `/wiki:compile --full` | Recompile everything from scratch |
  532. | `/wiki:query <question>` | Q&A against the wiki (standard depth) |
  533. | `/wiki:query <question> --quick` | Fast answer from indexes only |
  534. | `/wiki:query <question> --deep` | Thorough — reads everything, checks raw + sibling wikis |
  535. | `/wiki:query <question> --include-archived` | Explicitly search/read archived material, with archived citations labeled |
  536. | `/wiki:query <terms> --list` | Find content by keyword, tag, or category (replaces old `/wiki:search`) |
  537. | `/wiki:query --resume` | Reload context after a session break — recent activity, stats, last-updated articles |
  538. | `/wiki:plan <goal>` | Generate wiki-grounded implementation plan (interview → gap research → phased plan) |
  539. | `/wiki:plan <goal> --quick` | Plan from wiki content only — skip interview and gap research |
  540. | `/wiki:plan <goal> --format rfc\|adr\|spec` | Output as RFC, ADR, or tech spec instead of roadmap |
  541. | `/wiki:research <topic>` | 5 parallel agents: academic, technical, applied, news, contrarian |
  542. | `/wiki:research <topic> --new-topic` | Create a topic wiki and start researching — works from any directory |
  543. | `/wiki:research <topic> --min-time 1h` | Keep researching in rounds until time budget is spent |
  544. | `/wiki:research <topic> --plan` | Decompose into 3-5 parallel paths, confirm, then dispatch all at once |
  545. | `/wiki:research <topic> --specialist <name>` | Apply an enabled personal specialist method; repeat at most three times |
  546. | `/wiki:research <topic> --no-specialists` | Run the baseline workflow without loading specialist bodies |
  547. | `/wiki:research <topic> --deep` | 8 agents: adds historical, adjacent, data/stats |
  548. | `/wiki:research <topic> --retardmax` | 10 agents: skip planning, max speed, ingest aggressively |
  549. | `/wiki:thesis <claim>` | Thesis-driven research: evidence for + against → verdict |
  550. | `/wiki:thesis <claim> --min-time 1h` | Multi-round thesis investigation with anti-confirmation-bias |
  551. | `/wiki:lint` | Run health checks on the wiki |
  552. | `/wiki:lint --fix` | Auto-fix structural issues |
  553. | `/wiki:lint --deep` | Web-verify facts and suggest improvements |
  554. | `scripts/llm-wiki lint [path] --fail-on <level> --strict-engine` | Gate deterministic lint on severity and optional `config.md` engine declarations |
  555. | `scripts/llm-wiki identity status\|assign [path] [--apply]` | Inspect or deliberately assign a stable `wiki_id` shared with the registry |
  556. | `/wiki:audit` | Umbrella trust audit: wiki, outputs, provenance, and fresh research when needed |
  557. | `/wiki:audit --artifact <path>` | Audit one article or output artifact and follow its evidence chain |
  558. | `/wiki:audit --project <slug>` | Audit one project's outputs and upstream wiki state |
  559. | `/wiki:audit report` | Display the latest umbrella audit report |
  560. | `/wiki:librarian` | Focused wiki maintenance: staleness and quality scan for the `wiki/` layer |
  561. | `/wiki:librarian --article <path>` | Scan a single article |
  562. | `/wiki:librarian scan --passes staleness,quality,conventions` | Generate topic-guide advice/proposals for established wikis without rewriting `schema.md` |
  563. | `/wiki:librarian report` | Display the latest librarian scan report |
  564. | `/wiki:refresh [<article-path>|--due]` | Re-check article sources for changed facts and offer human-gated updates |
  565. | `/wiki:output <type>` | Generate: summary, report, study-guide, slides, timeline, glossary, comparison |
  566. | `/wiki:output <type> --retardmax` | Ship it now — rough but comprehensive, iterate later |
  567. | `/wiki:checkpoint create [<slug>] --repo <path> --audience private\|team\|public` | Preview a comprehensive cross-topic checkpoint with source access, section coverage, detail-floor, privacy, and gap review |
  568. | `/wiki:checkpoint create ... --apply` | Stage, synthesize, deterministically privacy-seal, verify, and write a five-file repo handoff without committing or pushing |
  569. | `/wiki:checkpoint refresh <bundle> [--apply]` | Re-run stored scope and review source/claim/privacy differences before replacing generated files |
  570. | `/wiki:checkpoint verify <bundle>` | Recheck structure, checkpoint/file hashes, privacy findings, and exact recorded overrides |
  571. | `/wiki:checkpoint import <bundle\|repo> --wiki <name>` | Verify and ingest a pinned checkpoint as bounded source evidence, never canonical truth |
  572. | `/wiki:project new <slug> "goal"` | Group related outputs under `output/projects/<slug>/` with a plain `WHY.md` rationale |
  573. | `/wiki:project list\|show\|add\|archive` | Manage output project folders without duplicating project state into frontmatter |
  574. | `/wiki:retract [<source-path>]` | User-authoritative removal of a source and its downstream wiki references, including selected archives and sessions |
  575. | `scripts/llm-wiki retract --everywhere [--apply] [--remove-from-logs]` | Secret-safe local value scan: hidden/stdin input, common encoded forms, optional whole-entry log removal, then verification |
  576. | `/wiki:ll` | Extract lessons learned from the current session into the wiki |
  577. | `/wiki:ll --dry-run` | Preview extracted lessons without writing |
  578. | `/wiki:ll --rules` | Also suggest CLAUDE.md / AGENTS.md rule additions |
  579. | `/wiki:assess <path>` | Assess a repo against wiki research + market. Gap analysis. |
  580. | `/wiki:assess <path> --retardmax` | Wide net — adds adjacent fields and failure analysis |
  581. All commands accept `--wiki <name>` to target a specific topic wiki and `--local` to target the project wiki. Archived topic wikis are skipped by default; commands that support `--include-archived` require that explicit flag before reading or writing archived material. Retraction is the exception: an explicitly selected archived target needs no additional gate, and `scripts/llm-wiki retract --everywhere` includes archives and session context. Commands that generate content (`query`, `output`, `plan`) also accept `--with <wiki>` to load supplementary wikis as cross-wiki context — e.g., `--with article-writing` applies writing craft knowledge when generating output from a domain wiki.
  582. `/wiki:librarian` is the focused wiki-maintenance tool. `/wiki:audit` is broader and may perform fresh research to decide whether the current knowledge or generated outputs are still trustworthy.
  583. ## How It Works
  584. ### Architecture
  585. ```
  586. ~/wiki/ # Hub — lightweight, no content
  587. ├── wikis.json # Registry of all topic wikis
  588. ├── _index.md # Lists topic wikis with stats
  589. ├── log.md # Global activity log
  590. ├── .sessions/ # Optional session capture + feedback candidates
  591. ├── .skills/ # Optional personal specialist methods + topic allowlists
  592. └── topics/ # Each topic is an isolated wiki
  593. ├── nutrition/ # Example topic wiki
  594. │ ├── .obsidian/ # Optional Obsidian vault config
  595. │ ├── inbox/ # Drop zone for this topic
  596. │ ├── inventory/ # Lazy: durable records, Ideas, and derived views
  597. │ ├── datasets/ # Lazy: manifests for large/external data
  598. │ ├── raw/ # Immutable sources
  599. │ ├── wiki/ # Compiled articles
  600. │ │ ├── concepts/
  601. │ │ ├── topics/
  602. │ │ └── references/
  603. │ ├── output/ # Generated artifacts
  604. │ ├── _index.md
  605. │ ├── config.md
  606. │ ├── schema.md # Human-owned topic guide
  607. │ └── log.md
  608. ├── woodworking/ # Another topic wiki
  609. ├── .archive/ # Archived topic wikis, hidden by default
  610. └── ...
  611. ```
  612. The hub has no content directories or `.obsidian/`; optional `.sessions/` and
  613. `.skills/` are operational configuration/memory layers. All content lives in
  614. topic sub-wikis with isolated indexes, articles, and a human-owned advisory
  615. topic guide (`schema.md`). Init creates the core wiki skeleton first; optional
  616. inventory and dataset layers are created when you use them. Queries stay
  617. focused. The multi-wiki peek finds overlap across topics when relevant.
  618. ### The Flow
  619. 1. **Research** a topic — parallel agents search the web, ingest sources, and compile articles in one command
  620. 2. **Apply specialist methods** — select a topic-allowlisted evidence/review protocol without pretending the model holds a credential
  621. 3. **Ingest** additional sources — URLs, files, text, tweets (via Grok MCP), or bulk via inbox
  622. 4. **Collect** catalogs of discoverable things — examples, media, memes, tools, projects, entities, or source candidates — then optionally inventory them
  623. 5. **Inventory** items, candidates, entities, corpora, watch lists, and next actions that should persist; the agent tells you when inventory is the wrong layer
  624. 6. **Develop Ideas** from rough seed to researched, challenged, approved shape, then explicitly promote them into delivery Projects
  625. 7. **Index datasets** that are too large for markdown — manifests, profiles, samples, and query recipes
  626. 8. **Archive** whole topic wikis that should stay preserved but quiet
  627. 9. **Compile** raw sources into synthesized wiki articles with cross-references and confidence scores
  628. 10. **Query** the wiki — quick (indexes), standard (articles), or deep (everything active, archived indexes separated)
  629. 11. **Session capture** — automatically preserve redacted Codex/Claude/OpenCode/Gemini checkpoints under `.sessions/` and rehydrate future turns
  630. 12. **Feedback curator** — capture reviewable correction/preference/approval candidates under `.sessions/feedback/` and promote only what matters
  631. 13. **Lessons learned** — extract knowledge from the current session (errors, fixes, gotchas) into the wiki
  632. 14. **Assess** a repo against the wiki — gap analysis: what aligns, what's missing, what the market offers
  633. 15. **Lint** for consistency — broken links, missing indexes, orphan articles, archive registry drift
  634. 16. **Output** artifacts — summaries, reports, slides — filed back into the wiki
  635. ### Key Design
  636. - **One topic, one wiki** — each research area gets its own sub-wiki with isolated indexes. No cross-topic noise.
  637. - **Canonical collection slugs** — collection families use kind-first topic
  638. names such as `memes-bitcoin`, `memes-ethereum`, or `tools-bitcoin` so
  639. related catalogs group naturally as the hub grows.
  640. - **Parallel research agents** — 5 standard, 8 deep, 10 retardmax. Each agent searches from a different angle.
  641. - **Personal specialist methods** — user-owned, instruction-only `SKILL.md`
  642. protocols selected by task features and enabled per topic; methods, not
  643. credential costumes.
  644. - **Collector workflow** — search-driven catalogs for objects, media, and
  645. examples; saves a provenance map first, then inventories only the durable
  646. subset.
  647. - **Concept → Idea → Project** — compiled Concepts hold knowledge, cataloged
  648. Ideas hold researched/shaped proposals, and Projects begin only after
  649. explicit delivery commitment. Natural language routes the whole lifecycle.
  650. - **Hub-wide portfolio** — one live, read-only view cross-references distributed
  651. Ideas and direct/promoted Projects without duplicating topic-owned records.
  652. - **`_index.md` navigation** — every existing wiki-managed directory has an index. Claude reads indexes first, never scans blindly.
  653. - **Articles are synthesized**, not copied — they explain, contextualize, cross-reference.
  654. - **Raw is immutable in normal workflows** — explicit user-directed retraction is the exception and removes selected raw and derived references.
  655. - **Typed raw namespaces** — keep high-volume streams under paths such as
  656. `raw/articles/news/daily/` without creating new types or losing per-directory
  657. indexes.
  658. - **Human-only verification** — compilation and due scans update operational
  659. dates, not `verified`; changed upstream material creates a new raw revision.
  660. - **Auditable lint identity** — reports include engine version/commit, optional
  661. corpus pins can gate CI, and stable wiki ids prevent registry paths from
  662. silently resolving to the wrong corpus.
  663. - **Multi-wiki aware** — queries peek at sibling wiki indexes for overlap.
  664. - **Archive-aware** — archived topic wikis stay preserved under
  665. `topics/.archive/` but are hidden from default query/compile/research/collect/output
  666. and maintenance workflows.
  667. - **Dual-linking** — both `[[wikilinks]]` (Obsidian) and standard markdown links on every cross-reference. Works everywhere.
  668. - **Confidence scoring** — articles rated high/medium/low based on source quality and corroboration.
  669. - **Structural guardian** — checks indexes, provenance, footnotes, wikilinks,
  670. and raw coverage; it auto-attaches a pointer only when one exact compiled
  671. owner exists and leaves editorial ambiguity alone.
  672. - **Activity log** — `log.md` tracks every operation, append-only and grep-friendly, with an explicit whole-entry privacy-retraction option.
  673. - **Opinionated inventory** — durable tracking gets records; one-off sources stay
  674. ingest/query; large row-like data becomes datasets or collection ingests. Big
  675. pivots start with a sample table before records are written.
  676. - **Media-safe catalogs** — bounded public binary media is cached under
  677. `output/assets/collect-<slug>/` by default; raw sources remain textual
  678. evidence, and large media sets become dataset manifests. Bulk downloads use
  679. timeouts, file-size caps, content-type checks, and IPv4 retry when needed.
  680. - **Feedback is candidate memory** — corrections, preferences, approvals, and
  681. plan acceptance are saved as redacted candidates under `.sessions/feedback/`;
  682. generic acknowledgements are ignored, and promotion into topic notes is
  683. explicit.
  684. - **Zero dependencies** — runs entirely on built-in tools (Claude Code, OpenCode, or Codex).
  685. ## Research Modes
  686. | Mode | Flag | Agents | Style |
  687. |------|------|--------|-------|
  688. | Standard | *(default)* | 5 | Academic, technical, applied, news, contrarian |
  689. | Deep | `--deep` | 8 | Adds historical, adjacent fields, data/stats |
  690. | Retardmax | `--retardmax` | 10 | Adds rabbit-hole agents. Skip planning, cast widest net, ingest aggressively, compile fast. Lint later. |
  691. **Smart input detection** — `/wiki:research` auto-detects whether you're passing a topic or a question:
  692. | Input | Detected as | Behavior |
  693. |-------|-------------|----------|
  694. | `"nutrition"` | Topic | Standard research — explore the field |
  695. | `"What makes articles go viral?"` | Question | Decompose into sub-questions → one agent per sub-question → synthesize → generate playbook → suggest theses |
  696. Question mode produces a **playbook** (actionable output artifact) and suggests **testable theses** derived from the findings.
  697. **Modifiers** (combine with any mode):
  698. | Flag | What it does |
  699. |------|-------------|
  700. | `--new-topic` | Create a topic wiki from the research topic and start immediately. Works from any directory. |
  701. | `--plan` | Decompose into 3-5 parallel research paths, confirm, then dispatch all paths simultaneously. Parallel ingest, sequential compile. |
  702. | `--min-time <duration>` | Keep running research rounds until the time budget is spent (`30m`, `1h`, `2h`, `4h`). Each round drills into gaps the previous round found. |
  703. | `--sources <N>` | Sources per round (default: 5, retardmax: 15) |
  704. ```
  705. # The full combo — new topic, 2 hours of deep research, from anywhere
  706. /wiki:research "CRISPR gene therapy" --new-topic --deep --min-time 2h
  707. ```
  708. Retardmax mode is inspired by [Elisha Long's retardmaxxing philosophy](https://www.retardmaxx.com/) — act first, think later. The antidote to analysis paralysis. Works for both `/wiki:research` and `/wiki:output`.
  709. ## Thesis-Driven Research
  710. Unlike open-ended research, `/wiki:thesis` starts with a specific claim and evaluates it:
  711. ```
  712. /wiki:thesis "intermittent fasting reduces neuroinflammation via glymphatic upregulation"
  713. ```
  714. **How it works:**
  715. 1. Decomposes the thesis into key variables, testable predictions, and falsification criteria
  716. 2. Launches parallel agents — but each agent has the thesis as a FILTER. Irrelevant sources get skipped (this prevents bloat)
  717. 3. Agents are split: **supporting**, **opposing**, **mechanistic**, **meta/review**, **adjacent** — balanced by design
  718. 4. Compiles evidence into wiki articles + a thesis file with evidence tables
  719. 5. Delivers a **verdict**: supported / partially supported / contradicted / insufficient evidence / mixed
  720. **Anti-confirmation-bias**: When using `--min-time`, Round 2 automatically focuses harder on the WEAKER side of the evidence. If Round 1 found mostly supporting evidence, Round 2 hunts for counter-evidence.
  721. **The thesis is the bloat filter.** Sources that don't relate to the claim's variables don't get ingested. Higher skip rate = tighter focus.
  722. ## Linking: Works Everywhere
  723. Every cross-reference in the wiki uses dual-link format:
  724. ```markdown
  725. [[gut-brain-axis|Gut-Brain Axis]] ([Gut-Brain Axis](../concepts/gut-brain-axis.md))
  726. ```
  727. The wiki is **not locked into any tool**:
  728. - **Obsidian** reads the `[[wikilink]]` — graph view, backlinks panel, quick-open
  729. - **Claude Code** follows the standard `(relative/path.md)` link
  730. - **GitHub/any markdown viewer** renders the standard link as clickable
  731. - **No viewer at all** — plain markdown, readable in any text editor
  732. ## Obsidian Integration
  733. Each topic wiki has its own `.obsidian/` config and can be opened as an independent vault:
  734. ```
  735. open ~/wiki/topics/nutrition/ # Open in Obsidian — focused graph for one topic
  736. ```
  737. The hub (`~/wiki/`) has no `.obsidian/` to avoid nested vault confusion. If you want a cross-topic view, open `~/wiki/` manually and let Obsidian create its own config.
  738. What works out of the box:
  739. - `.obsidian/` config can be created on init with sane defaults
  740. - `[[wikilinks]]` power the graph view
  741. - `aliases` in frontmatter enable search by alternate names
  742. - `tags` in frontmatter are natively read
  743. - `inbox/` works as a drop zone in both Obsidian and the CLI
  744. Claude Code is the compiler. Obsidian is an optional viewer.
  745. ## Query Depths
  746. | Depth | Flag | What it does |
  747. |-------|------|-------------|
  748. | Quick | `--quick` | Reads indexes only. Fastest. For simple lookups. |
  749. | Standard | *(default)* | Reads relevant articles + full-text search. For most questions. |
  750. | Deep | `--deep` | Reads everything active, searches raw sources, peeks sibling wikis, and surfaces archived index matches separately. |
  751. | List | `--list` | Returns ranked article list instead of synthesized answer. Supports `--tag` and `--category` filters. |
  752. Archived topics are excluded from quick, standard, and list results unless you
  753. pass `--include-archived`. Deep mode may show archived index hits, but it does
  754. not cite archived material as active evidence without explicit inclusion.
  755. ## Development Benchmarks
  756. Token-efficiency changes are tested with deterministic context budgets and
  757. optional real Codex and Claude Code AB/BA harnesses. See
  758. [`benchmarks/README.md`](benchmarks/README.md) for commands, quality gates, and
  759. cache-metric caveats.
  760. ## Credits
  761. - [Andrej Karpathy](https://x.com/karpathy) — the [LLM wiki concept](https://x.com/karpathy/status/2039805659525644595) and [idea file](https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f)
  762. - [Elisha Long](https://www.retardmaxx.com/) — retardmaxxing philosophy (act first, think later)
  763. - [tobi/qmd](https://github.com/tobi/qmd) — recommended local search engine for scaling beyond ~100 articles
  764. - [rvk7895/llm-knowledge-bases](https://github.com/rvk7895/llm-knowledge-bases) — prior art in Claude Code wiki plugins
  765. ## License
  766. MIT License. Copyright (c) 2026 nvk.
  767. Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files, to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, subject to the following conditions: the above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
  768. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.