Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add commands/nvk/llm-wiki/adaptergit clone --depth 1 https://github.com/nvk/llm-wikiWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00027 | $0.00967 |
| Opus 5 | $0.00014 | $0.00483 |
| Sonnet 5 | $0.00005 | $0.00193 |
| Haiku 4.5 | $0.00003 | $0.00097 |
Grade A, and why
adapter scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 3d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 87 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Your task
Manage or invoke a private llm-wiki adapter through the deterministic bundled
CLI. Read skills/wiki-manager/references/adapters.md before acting.
Private-adapter management is wiki-neutral. Do not put executable registrations
or absolute machine paths in wikis.json or a topic wiki. Resolve a topic wiki
only when the user asks to promote a reviewed result after execution.
Declarative intent routing
Before treating an external URL as an ingestion source, normalize the requested effect to a lowercase intent token and run:
${CLAUDE_PLUGIN_ROOT}/bin/llm-wiki adapter route \
--intent <effect> --resource '<external-url>' --json
On matched, read the returned adapter-owned guide, then run show and
doctor for the returned adapter id before following that workflow. The private
adapter owns all provider-specific authentication, transport, planning,
recovery, and verification instructions. On no-match, continue with normal
wiki routing. On ambiguous or unavailable, fail closed and repair the
registration instead of guessing an adapter or falling back to an ungoverned
write path. Route output does not echo the resource.
A URL by itself never authorizes invented edits. A bounded imperative approves only a faithful plan; generic remote-write plan hashes, revision locks, idempotency keys, private receipts, and read-back verification remain required.
Locate the CLI
For Claude Code, use:
"${CLAUDE_PLUGIN_ROOT}/bin/llm-wiki" adapter <subcommand>
In a source checkout, use scripts/llm-wiki. Other runtimes should use the
bundled bin/llm-wiki relative to the installed plugin root described in the
adapter reference. Do not assume the command is globally installed.
Management operations
add <path>: require an existing local checkout and explicit read/write roots. Add exact--remote-resourcevalues only when the user authorized them. Never clone a repo or request credentials. Use--replaceonly when the user intends to trust a changed manifest.list: show compact id, version, capability, network, and local-root data.show <id>: show the machine-local registration; do not print environment variable values.route --intent <effect> --resource <url>: match only registered declarative routes and return the adapter-owned guide without echoing the resource.doctor <id>: verify manifest hash, executable, and handshake before a run.remove <id> --yes: remove only the registration. Never delete adapter code, inputs, or outputs.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 3d ago First seen · 87 lines · 27 tokens per session scan A 7cee33f5250e
adapter is a command published in the GitHub repository nvk/llm-wiki (1,175 stars, last pushed 6d ago), licensed MIT. It adds 27 tokens to every session and 967 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other commands, from other repositories
workflow-claude-commands
Track Claude Code commands report changes and find what needs updating.
workflow-claude-subagents
Track Claude Code subagents report changes and find what needs updating.
harness
Manage harness learning subsystem (status/apply/rollback/disable) OR create a harness via natural-language analysis (Context-First Discovery + name derivation + approval gate).
become
Agent crystallization command. Studies a tool, role, or person and produces a portable 9-section agent that inhabits the domain - with beliefs, scar tissue, and instincts.
reflect
Apply captured learnings to improve the Craft harness and project patterns. Converts .learnings.yaml into permanent updates.
ask
Consult a craft agent. Routes your question to the best mind in the workshop - not a menu, a recommendation.