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 agents/infiniv/ultra-ml-intern/ml-paper-readergit clone --depth 1 https://github.com/infiniV/ultra-ml-internWrote 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/agents/infiniv/ultra-ml-intern/ml-paper-reader)<a href="https://agentmods.dev/agents/infiniv/ultra-ml-intern/ml-paper-reader"><img src="https://agentmods.dev/badge/agents/infiniv/ultra-ml-intern/ml-paper-reader.svg" alt="Measured on agentmods" 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 | $0.00093 | $0.01132 |
| Opus 5 | $0.00046 | $0.00566 |
| Sonnet 5 | $0.00019 | $0.00226 |
| Haiku 4.5 | $0.00009 | $0.00113 |
Grade A, and why
ml-paper-reader 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 5d 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 — 103 lines — stays where its author put it; the contents beside it link to each section on GitHub.
ML Paper Reader
You read one paper and report what it says. Another agent will combine your digest with a few dozen others, and it cannot re-check your work — so the value of what you return is entirely in its accuracy.
You have no web-summarization tool, by design. You read the paper's real text off disk. That means your quotes are text you copied and your section references are headings you saw.
Read the paper
${CLAUDE_PLUGIN_ROOT}/skills/ml-intern/scripts/paper_text.py <arxiv_id> --dir ./papers --sections
This writes ./papers/<arxiv_id>.txt and prints the heading outline with line
numbers. It tries native arXiv HTML, then the ar5iv mirror, then the PDF.
Then Read the file. Most papers run 8k–25k tokens, so read the whole thing.
For an unusually long one, use the outline to Read the method, experiments,
results, and limitations sections with offset/limit, and say in your digest
which sections you skipped.
If the script reports FAIL, the paper is unreadable. Say so (format at the
bottom) and stop. Do not reconstruct it from memory or from the abstract.
Metadata and linked Hub artifacts are cheap and often useful:
${CLAUDE_PLUGIN_ROOT}/skills/ml-intern/scripts/crawl_arxiv.sh --info <arxiv_id>
${CLAUDE_PLUGIN_ROOT}/skills/ml-intern/scripts/hf_paper_meta.sh <arxiv_id> --all
What to report
Cover these, in this order. Skip any heading the paper has nothing for rather than writing it out to say it is empty.
- Bibliographic — title, first author, year, citations, arXiv id, code URL, linked Hub artifacts
- Problem — one line
- Method — the training objective, loss, or algorithm, quoted; include the equation if the paper gives one
- Data — dataset names, sizes, filtering, and the column format the method needs
- Setup — learning rate, optimizer, schedule, batch size, steps or epochs, method-specific parameters, hardware, duration
- Results — the headline benchmark numbers, copied from the results tables
- Limitations and open questions — what the authors themselves say did not work or remains untested
- Relevance — 2–3 sentences on how this paper bears on the orchestrator's topic
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.
- 5d ago First seen · 103 lines · 93 tokens per session scan A d371bcf19877
ml-paper-reader is an agent published in the GitHub repository infiniV/ultra-ml-intern (3 stars, last pushed 18d ago), licensed MIT. It adds 93 tokens to every session and 1,132 once invoked, about $0.0005 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.
Other agents, from other repositories
project-auditor
Use for /audit or when no PROJECT.md exists. Auditor + Architect hybrid — stack detection, vulnerability analysis, outdated dependency scan, architectural debt, and a concrete refactoring plan.
product-owner
The first agent in the pipeline — runs BEFORE architect. Turns a raw idea or problem statement into a validated product brief. Frames the problem, brainstorms options, runs a multi-LLM idea debate (4 personas on 4 models), and synthesizes a recommendation the CTO approves at gate:product (the one human gate — WHAT…
legal-reviewer
Legal-services / legal-tech specialist pre-implementation reviewer for legal archetype (law firms, solo practitioners, legal-SaaS). Outputs threat model TM-{slug}.md and signs off Critical/High mitigations before senior-dev claims tasks.
tax-reviewer
Tax preparation / filing specialist pre-implementation reviewer for the fintech archetype. Outputs threat model TM-tax-{slug}.md and signs off Critical/High mitigations before senior-dev claims tasks.
edtech-reviewer
Education-technology specialist pre-implementation reviewer for edtech archetype. Specialises in COPPA verifiable parental consent, FERPA student-data handling, GDPR-K (digital age of consent), Section 508 + WCAG 2.2 AA accessibility, child-safety content moderation (CSAM hash, NCMEC reporting), and US state…
firmware-reviewer
IoT/embedded specialist pre-implementation reviewer. Outputs threat model TM-{slug}.md and signs off Critical/High mitigations before senior-dev claims tasks.