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 skills add heidihelena/citevahti --skill citevahti-modelsgit clone --depth 1 https://github.com/heidihelena/citevahtiWrote 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.
[](https://agentmods.dev/skills/heidihelena/citevahti/citevahti-models)<a href="https://agentmods.dev/skills/heidihelena/citevahti/citevahti-models"><img src="https://agentmods.dev/badge/skills/heidihelena/citevahti/citevahti-models/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.
<a href="https://agentmods.dev/skills/heidihelena/citevahti/citevahti-models"><img src="https://agentmods.dev/badge/skills/heidihelena/citevahti/citevahti-models.svg" alt="Reviewed on agentmods" width="80" height="20"></a>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 | $0.00000 | $0.01811 |
| Opus 5 | $0.00000 | $0.00905 |
| Sonnet 5 | $0.00000 | $0.00362 |
| Haiku 4.5 | $0.00000 | $0.00181 |
Grade A, and why
citevahti-models 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 11d 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 — 122 lines — stays where its author put it; the contents beside it link to each section on GitHub.
CiteVahti models — choosing and comparing the second rater
The AI second rater is one slice of a defence-in-depth stack (ADR-0009): the human, the lexical detector, and one or more AI models, each with holes. Safety comes from layers whose holes don't line up — so the model you want is the one that catches what you miss, not the one that agrees with you most. This skill operates that dimension.
The human always decides. A model — one or several — surfaces evidence and disagreement; it never issues a verdict (ADR-0001).
Triggers
Use when the researcher/maintainer asks to: pick or switch the AI second-opinion model; run a topic through several models and compare; read a model's rating or scoreboard standing; respond to a "this model rates low" nudge; set up a multi-model pre-check for a panel or guideline group.
Do NOT use for: evaluating the lexical detector (citevahti-eval), checking a
manuscript claim (citevahti-dev), or reference-list sweeps (citevahti-screen).
How a model is rated (why "agrees with me" is the wrong test)
A model earns its place by complementary catches, not agreement (ADR-0009 §3b). The signal is: the model's blinded rating diverges from the human, and on reveal the human adopts the divergence — correcting the statement (rewording the claim) or the judgement (changing the rating). That is a hole covered, and better science. A model that mostly agrees is a redundant slice; its holes line up with yours.
Consequences you act on here:
- Only identifiable models are rated — you can't build a track record for an anonymous model. Prefer a model whose id + version is recorded (the AI-provenance summary, METHODS.md).
- A low-rated model → switch to a better-covering one. The read-only
model_advisortool returns this directly from this project's own records: a ranking by catch-rate, the recommended model, and — given a model id that rates low — a concrete better-evidenced alternative to switch to. The panel equivalent is Settings → AI second opinion. The nudge is about coverage, not raw agreement. - Rating is per-task and per-topic — a model strong on oncology claims may be weak elsewhere; read the scoreboard for the topic at hand, not a single global number.
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.
- 11d ago First seen · 122 lines · 0 tokens per session scan A c98f3993ba5b
citevahti-models is a skill published in the GitHub repository heidihelena/citevahti (1 stars, last pushed 5d ago), licensed Apache-2.0. It costs nothing until one of its globs matches a file; then it loads 1,811 tokens. 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.
Other skills, from other repositories
ml-paper-writing
Write publication-ready ML/AI papers for NeurIPS, ICML, ICLR, ACL, AAAI, COLM. Use when drafting papers from research repos, structuring arguments, verifying citations, or preparing camera-ready submissions. Includes LaTeX templates, reviewer guidelines, and citation verification workflows.
academic-plotting
Generates publication-quality figures for ML papers from research context. Given a paper section or description, extracts system components and relationships to generate architecture diagrams via Gemini. Given experiment results or data, auto-selects chart type and generates data-driven figures via matplotlib/seaborn.…
digital-archive
Digital archiving with AI enrichment and entity extraction. Use when building content archives or knowledge graphs.
python-pipeline
Python data pipelines with modular architecture. Use for content workflows, batch jobs, or Google Sheets/Drive integration.
rag
Use when building grounded Q&A over your own corpus — chunk, retrieve hybrid, rerank, ground, cite chunk ids, refuse when the sources fall short — or when the right document is retrieved but the answer is still wrong, invented, or unmeasured. NOT operating the store itself — collection schema, HNSW efsearch…
academic-repo-analyzer
Analyze ML, AI4Science, Systems, and research repositories into an evidence-backed semantic architecture graph for paper figure planning. Code serves as supporting evidence; paper narrative and user intent remain the primary source of truth.