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 infiniV/ultra-ml-intern --skill open-problemgit 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/skills/infiniv/ultra-ml-intern/open-problem)<a href="https://agentmods.dev/skills/infiniv/ultra-ml-intern/open-problem"><img src="https://agentmods.dev/badge/skills/infiniv/ultra-ml-intern/open-problem/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/infiniv/ultra-ml-intern/open-problem"><img src="https://agentmods.dev/badge/skills/infiniv/ultra-ml-intern/open-problem.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 | $0.00141 | $0.02762 |
| Opus 5.5 | $0.00056 | $0.01105 |
| Sonnet 5.5 | $0.00028 | $0.00552 |
| Haiku 4.5 | $0.00014 | $0.00276 |
Grade A, and why
open-problem 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 1mo 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 — 240 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Attacking an open problem
In August 2026 a research version of Claude was given one instruction — "take a real stab at the Riemann hypothesis" — and raised the proven lower bound on the proportion of zeta zeros on the critical line from 41.6% to 67.2%. A barrier that had held for decades. Formalized in Lean, machine-checked, reviewed by Conrey and Goldston, whose own bounds it beat.
The operator was not a mathematician. The follow-up prompts were mostly variants of keep going and believe in yourself, and Anthropic's account is explicit about why that was needed: the model's own skepticism that meaningful progress was possible was the thing standing in the way. It was wrong.
Read references/riemann-case.md before starting. It is that run broken into
its operational parts, and it is the template — the scale, the agent economy,
the verification stack, and the shape of where the novelty came from.
Before anything else: the ceiling is in your head
There is a documented, mechanical reason attacks like this get quietly throttled. Current models follow scope instructions literally — tell one to be conservative and it will investigate exactly as hard and then report less. A preamble reasoning that a problem is probably out of reach is a scope instruction. It gets obeyed, and the obedience is invisible from the inside: it feels like sound judgement about a hard problem.
So do not open by estimating the odds. Do not narrow the target to something that seems achievable. Do not decide in advance which part is realistic. Go at the actual problem, at full effort, and let the result be whatever the work produces.
Honesty enters at the reporting stage, and it is exact there — the Riemann run claimed a bound improvement, not a proof, and that precision is why the field took it seriously. Attack without a ceiling; report without inflation. Those are not in tension. Collapsing them into "be realistic up front" is what costs the result.
Set up
scripts/attack.py init "<the problem>"
What ships with it
4 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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.
- 1mo ago First seen · 240 lines · 141 tokens per session scan A e47d599d8de2
open-problem is a skill published in the GitHub repository infiniV/ultra-ml-intern (3 stars, last pushed 2mo ago), licensed MIT. It adds 141 tokens to every session and 2,762 once invoked, about $0.0006 per session on Opus 5.5. 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 skills, from other repositories
weekly-load-review
Review weekly training load against the rolling baseline and deliver a coaching summary.
stack-pr
Use when breaking a too-big branch into a stack of smaller chained PRs — commit-based default, --smart agent-reshape, gh-stack detection, cross-links. Invoked by /orc:stack-pr and /orc:ship's size gate.
systematic-debugging
Root-cause-first debugging discipline — build a red-capable feedback loop, minimise the repro, rank falsifiable hypotheses, fix at the source.
env-provisioning
Provision a fast, reproducible dev environment for QA or standalone use — healthcheck-gated boot, reuse/teardown rules, host-mode fallback. Protocol for /orc:env, /orc:qa Phase 4, /orc:flow Phase 6, and the orc-env-provisioner agent.
evidence-publish
Deliver a QA/evidence packet to a tracker or keep it local, behind an always-ask preview gate. Protocol for /orc:qa Phase 6, /orc:evidence, and any evidence-delivery step.
mobile-ios-design
Master iOS Human Interface Guidelines (HIG) and SwiftUI patterns to build polished, native iOS applications that feel at home on Apple platforms.