A benchmark for measuring how well a selected language model or agent command-line tool creates a cited wiki from a source collection. Unlike a pipeline test, it compares the model while keeping the wiki-building process fixed.

Use it to run model comparisons, grade wiki-building results, inspect structural checks and citations, and analyze errors such as missed updates, contradictions, or false claims.

It shows whether differences in wiki quality come from the chosen model, including missed contradictions or incorrect claims, rather than from changes to the processing pipeline.

How MarkusNeusinger/cite-citadel configures Claude Code

Reuse

Borrowing it

Nothing to install: this file belongs to MarkusNeusinger/cite-citadel. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.

Copy the file

curl -O https://raw.githubusercontent.com/MarkusNeusinger/cite-citadel/main/.claude/skills/bench-model/SKILL.md

Clone the repo

git clone --depth 1 https://github.com/MarkusNeusinger/cite-citadel

Made for: Claude Code.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

README.md

[![agentmods](https://agentmods.dev/badge/skills/markusneusinger/cite-citadel/bench-model/github.svg)](https://agentmods.dev/skills/markusneusinger/cite-citadel/bench-model)

Your own site

<a href="https://agentmods.dev/skills/markusneusinger/cite-citadel/bench-model"><img src="https://agentmods.dev/badge/skills/markusneusinger/cite-citadel/bench-model/github.svg" alt="Measured on agentmods" height="20"></a>

Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.

Your own site · 80×15

<a href="https://agentmods.dev/skills/markusneusinger/cite-citadel/bench-model"><img src="https://agentmods.dev/badge/skills/markusneusinger/cite-citadel/bench-model.svg" alt="Reviewed on agentmods" width="80" height="20"></a>

Per session 306 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.

When invoked 2,536 The whole file, excluding the scripts and references it only reads on demand.

Security scan C 1 finding. A grade says what 26 rules found in the file — not that it is safe.

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.

Model	Per session	Once invoked
Fable 5.1	$0.00306	$0.02536
Opus 5.5	$0.00122	$0.01014
Sonnet 5.5	$0.00061	$0.00507
Haiku 4.5	$0.00031	$0.00254

Measured 2mo ago against content hash 54e4ffde665a, method: parsed. Prices are Anthropic first-party input rates as of 2026-10-07, from the pricing page.

Security

Grade C, and why

bench-model scanned grade C with 1 finding 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 2mo 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.

Recursive force deletehighDestructive command

rm -rf with a variable or a broad path is one typo away from removing the wrong tree.

throwaways (`rm -rf`), kept only while a `--baseline` comparison still needs them.

.claude/skills/bench-model/SKILL.md · 149 lines

How it starts

The opening of the file, as written. The whole thing — 149 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Bench a model on wiki creation

Usage: bench-model <corpus> <model> [--cli claude|copilot|agy] [--baseline SANDBOX] [--grade-only SANDBOX]

Everything sandbox/grading-related follows verify-corpus (read its SKILL.md first — sandbox recipe, phase 1 structural gates, phase 2 retrieval-first grading, the creation-vs-retrieval miss matrix, the gotchas). This skill changes the question: not "does the pipeline still work?" but "how well does THIS model drive it?" — so the model is the variable, the grade gains a discriminative tier, and the output is a comparison, not just pass/fail.

Reference points (2026-07 audit, beverages, 14 sources)

	claude-sonnet-5	claude-haiku-4-5
pace	~7.5 min/source	~3 min/source (~2.5× faster)
structural gates	check+lint clean	check+lint clean
ground-truth grade	no wiki-defects on its subset	7 wiki-defects
judgment failures	none observed	3 uncaught contradictions; a planted-false claim ("coffee loses caffeine with age") adopted as unqualified wiki-voice truth; partial 2024→2026 supersession

The pattern to test for: cheap models produce structurally valid, well-cited pages at speed but miss judgment-heavy work — surfacing contradictions, quarantining suspicious claims, temporal supersession, open-points discipline. Structural gates alone will NOT separate models.

Mode A — build with the model under test

Per verify-corpus's sandbox recipe, with the model pinned; run ingest in the background (a foreground shell call gets killed by per-command timeouts long before a full corpus finishes):

REPO="$(git rev-parse --show-toplevel)"
SANDBOX="$(mktemp -d)/bench-<corpus>-<model>"
uv run python -m citadel init "$SANDBOX"
cat > "$SANDBOX/.env" <<EOF
CITADEL_LLM_CLI=claude                 # or copilot / agy — the seam for open/local models
CITADEL_INGEST_MODEL=<model>
CITADEL_LLM_LOG_DIR=$SANDBOX/logs
CITADEL_RAW_DIR=$REPO/corpora/<corpus>/raw
EOF
# sanity-check the model id first (cheap):  claude --model <model> -p "Reply with exactly: ok"
CITADEL_WORKSPACE="$SANDBOX" uv run python -m citadel ingest --quiet &   # background; poll:
uv run python -c "import json;print(len(json.load(open('$SANDBOX/wiki/.citadel_ingested.json'))['sources']))"

Read the full file on GitHub · 149 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.

2mo ago First seen · 149 lines · 306 tokens per session scan C 54e4ffde665a

Subscribe to this mod's changes

bench-model is a skill published in the GitHub repository MarkusNeusinger/cite-citadel (3 stars, last pushed 15d ago), licensed MIT. It adds 306 tokens to every session and 2,536 once invoked, about $0.0012 per session on Opus 5.5. A static security scan graded it C with 1 finding (recursive force delete). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.

Analytics and metrics Building agents agentic ai-tools citations claude cli knowledge-base

bench-model next to 6 skills from other repositories
Mod	Type	Stars	Last push	Licence	Scan	Tokens	Works with
bench-modelMarkusNeusinger/cite-citadel	Skill	3	15d ago	MIT	C	306	Claude Code
golang-manual-disoulcodex/agentic	Skill	10	7d ago	MIT	A	97	Claude Code, Codex
terraform-infrastructuresoulcodex/agentic	Skill	10	7d ago	MIT	A	96	Claude Code, Codex
git-flow-prsoulcodex/agentic	Skill	10	7d ago	MIT	A	103	Claude Code, Codex
github-actions-workflowsoulcodex/agentic	Skill	10	7d ago	MIT	A	72	Claude Code, Codex
microservices-architecturesoulcodex/agentic	Skill	10	7d ago	MIT	A	61	Claude Code, Codex
module-federationsoulcodex/agentic	Skill	10	7d ago	MIT	A	51	Claude Code, Codex

cite-citadel: Skill for Claude Code