llm-wiki

██╗     ██╗     ███╗   ███╗    ██╗    ██╗██╗██╗  ██╗██╗
██║     ██║     ████╗ ████║    ██║    ██║██║██║ ██╔╝██║
██║     ██║     ██╔████╔██║    ██║ █╗ ██║██║█████╔╝ ██║
██║     ██║     ██║╚██╔╝██║    ██║███╗██║██║██╔═██╗ ██║
███████╗███████╗██║ ╚═╝ ██║    ╚███╔███╔╝██║██║  ██╗██║
╚══════╝╚══════╝╚═╝     ╚═╝     ╚══╝╚══╝ ╚═╝╚═╝  ╚═╝╚═╝

github.com/nvk/llm-wiki

LLM-compiled knowledge bases for any AI agent. Parallel multi-agent research, collector catalogs, automated session capture, feedback curation, thesis-driven investigation, source ingestion, wiki compilation, truth-seeking audits, querying, and artifact generation. Ships as a Claude Code plugin, an OpenAI Codex plugin, an OpenCode instruction file, or a portable AGENTS.md for any other LLM agent. Obsidian-compatible.

Install · Quick Start · Commands · How It Works · Research Modes · Thesis Research · Query Depths · Linking · Obsidian · Architecture · Nono Sandbox · Upgrade · Changelog · Credits

Changelog

v0.12.0 — Feedback curator.

• Captures high-signal user corrections, preferences, approvals, and plan acceptance as redacted candidates under HUB/.sessions/feedback/.
• Ignores generic acknowledgements like ok, thanks, and cool.
• Adds @wiki feedback list|show|capture|promote for review, manual capture, and promotion.
• Promotes selected candidates explicitly into topic raw/notes/; nothing is auto-promoted.
• Includes recent feedback candidates in session digests for review during rehydration and lessons-learned workflows.

v0.11.1 — Session helper compatibility. Small hotfix so automated session capture works on Python 3.9/macOS system Python as well as newer Python runtimes.

v0.11.0 — Automated session capture.

• Default-on redacted session checkpoints under HUB/.sessions/.
• New @wiki session status|disable|enable|capture|list|show|rehydrate|promote workflow.
• session disable opt-out writes enabled: false, making trusted hooks no-op.
• Compact context rehydration for future turns/sessions.
• Captured digests stay operational until explicitly promoted into topic raw/notes/.
• Codex hook bundle plus tested Claude-shaped hook payload handling.

v0.10.2 — Collector production hardening. Collection-family topic slugs now prefer kind-first names such as memes-bitcoin, the scale boundary treats 500 rows as large and 501+ as huge, and media downloads call out timeouts, file-size caps, content-type checks, and IPv4 retry for hosts that hang.

v0.10.1 — Collector media downloads. /wiki:collect now downloads bounded public binary media into output/assets/collect-<slug>/ by default for media-bearing collections, records local paths and hashes in the catalog, and keeps --media reference as the no-download opt-out.

v0.10.0 — Collector catalogs. Added /wiki:collect for provenance-rich catalogs of examples, artifacts, media, memes, tools, entities, and source candidates. Collect infers scale, captures aliases and found-in-context provenance, handles binaries as referenced assets by default, writes output/collect-... catalogs, and promotes only selected durable subsets into inventory, raw sources, wiki articles, or datasets.

v0.9.0 — Topic archive lifecycle. Whole topic wikis can now be archived under topics/.archive/ so old interests stay preserved but out of normal context. Query, ingest, compile, research, output, inventory, datasets, projects, librarian, refresh, audit, lint, init, and routing now distinguish active material from explicitly included archived context.

v0.8.7 — iCloud permission diagnostics. When macOS lets Codex stat an iCloud hub path but denies actual reads or directory listings, the local CLI now reports the real privacy-permission problem instead of calling the registry invalid or suggesting a machine-local fallback path.

v0.8.6 — Lint repair correctness. lint --fix now repairs legacy article frontmatter, rewrites fuzzy raw source refs to exact paths when resolution is unambiguous, regenerates stale directory indexes, ignores maintenance backup indexes under .librarian/, and creates an explicit uncompiled-source coverage reference instead of leaving raw coverage gaps as endless suggestions.

v0.8.5 — Safer lint defaults. Hub-level lint now stays scoped to the shared registry instead of recursively auditing every topic by accident, and lint --fix preserves absent lazy inventory/ and datasets/ layers unless those layers already exist or the current workflow needs them.

v0.8.4 — Portable iCloud hub resolution. Shared wiki folders now survive moving between Macs with different /Users/<name>/... paths: agents prefer portable hub_path, treat legacy resolved_path values as fallback caches, resolve wikis.json paths relative to the current hub, and fall back to populated topics/<slug>/ directories when registry entries are stale or unreadable.

Install

Claude Code (native plugin):

claude plugin install wiki@llm-wiki

OpenAI Codex (marketplace plugin):

Install from GitHub:

codex plugin marketplace add nvk/llm-wiki
# Then open /plugins in Codex, enable "LLM Wiki", and use @wiki

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

Canonical explicit invocation:

@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

Remove:

codex plugin marketplace remove llm-wiki

Troubleshooting:

• After installing the marketplace, open /plugins in Codex and enable "LLM Wiki" — first install requires the interactive enable step.
• @wiki is the canonical explicit entry point in Codex. Natural-language wiki requests can still auto-activate the skill.
• 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 — Codex needs r+w to $HOME/.codex for plugin install.

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

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.

Web search requires export OPENCODE_ENABLE_EXA=1.

Pi (instruction file — best for local models):

Pi's minimal system prompt (~1K tokens) leaves room for the full wiki skill on 32K context local models.

pi --instructions path/to/llm-wiki/plugins/llm-wiki-opencode/skills/wiki-manager/SKILL.md

With a local llama-server backend (no cloud API needed):

OPENAI_BASE_URL=http://127.0.0.1:8080/v1 OPENAI_API_KEY=local \
  pi --instructions path/to/llm-wiki/plugins/llm-wiki-opencode/skills/wiki-manager/SKILL.md

Pi uses the same OpenCode skill file — no separate packaging needed.

Any LLM Agent (idea file):

# Copy AGENTS.md into your agent's context or project root
cp AGENTS.md ~/your-project/AGENTS.md

The AGENTS.md file contains the complete wiki protocol as a single portable document — works with any LLM agent that can read/write files and search the web.

Claude-First, Multi-Runtime

Claude Code is the principal user. Keep one shared behavior layer and thin packaging layers per runtime:

• claude-plugin/ is the primary distribution target and UX surface.
• claude-plugin/skills/wiki-manager/ is the behavioral source of truth.
• plugins/llm-wiki/skills/wiki/ is the generated Codex packaging target behind @wiki.
• plugins/llm-wiki-opencode/ is the OpenCode and Pi packaging target.
• .agents/plugins/marketplace.json makes the Codex plugin installable from this repo.
• AGENTS.md is the portable single-file protocol for any other LLM agent.

Supported clients:

Both runtime mirrors are generated, not hand-maintained. Rebuild from the Claude source of truth: