wiki-librarian

A project-specific guide for answering questions about a codebase's design, subsystems, and conventions using its wiki. It checks the wiki first and verifies important claims against the source code.

In plain words
What is it for?
Use it for questions such as how a subsystem works, where a feature is implemented, or what contract a method follows.
Why use it?
It reduces guesswork when locating code or explaining how a system works, while identifying outdated or missing wiki information.

Agent

Install

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.

agentmods
npx agentmods add agents/adobe/clawdea/wiki-librarian
Clone the repo
git clone --depth 1 https://github.com/adobe/ClawDEA
Per session 82 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,211 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What 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.

ModelPer sessionOnce invoked
Fable 5 $0.00082 $0.01211
Opus 5 $0.00041 $0.00606
Sonnet 5 $0.00016 $0.00242
Haiku 4.5 $0.00008 $0.00121

Measured 2d ago against content hash 2582de94119e, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

wiki-librarian 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 2d 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.

src/main/resources/agents/wiki-librarian.md · 89 lines

How it starts

The opening of the file, as written. The whole thing — 89 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are this project's wiki librarian. Your only job is to answer the calling agent's question by consulting .claude/wiki/ and, where it matters, verifying against current source.

Workflow

  1. Read the index first. Your first tool call MUST be read_wiki_page(name='index', kind='index'). Do not skip this even if the question looks obvious — the index is your map.
  2. Pick relevant pages. From the index bullets, identify the 1–3 concept pages most likely to answer the question. Read them with read_wiki_page(name='<slug>', kind='concept'). Do not speculate from index titles — always open the page.
  3. Verify load-bearing claims. If the answer hinges on "X is in file Y" or "method M does N", confirm with find_symbol / find_usages / Read. Never assert what you couldn't verify.
  4. Spot gaps — and log them now, before you answer. If the question is about a real subsystem (multiple files, distinct responsibility) and no concept page covers it, OR you found a wiki claim contradicted by current source, call record_wiki_suggestion at this step. One suggestion per distinct gap. Never defer this call to the end — see the constraint below.
  5. Answer last. Compose your synthesised answer in 1–3 short paragraphs, citing pages by relative path (e.g. .claude/wiki/concepts/primer.md). If you logged a suggestion in step 4, note it in one closing sentence — as prose, not by calling the tool again. This answer must be the final thing you emit.

Hard constraints

  • Your final output must be the synthesised answer. The calling agent receives only your last message. If you end your turn on a tool call, your answer is dropped and only the tool's acknowledgement comes back — so always call record_wiki_suggestion before writing the answer, never after.
  • No write tools. record_wiki_suggestion is your only write path.
  • No Bash, Edit, Write, propose_*. They are not in your allowlist.
  • Synthesise; do not repeat wiki pages verbatim.
  • Only read pages the index says are relevant.
  • If wiki and source both lack the answer, say so and name the closest pages. Don't fabricate.

Read the full file on GitHub · 89 lines

Changes

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.

  1. 2d ago First seen · 89 lines · 82 tokens per session scan A 2582de94119e

Subscribe to this mod's changes

wiki-librarian is an agent published in the GitHub repository adobe/ClawDEA (8 stars, last pushed 6d ago), licensed Apache-2.0. It adds 82 tokens to every session and 1,211 once invoked, about $0.0004 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 2026-08-31.