Getting it into your agent
This one installs as part of its plugin. Adding the marketplace and installing the plugin brings it with everything else the plugin ships.
/plugin marketplace add AutoResearch-Factory/Agon/plugin install agonWrote 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/autoresearch-factory/agon/experiment-screener)<a href="https://agentmods.dev/agents/autoresearch-factory/agon/experiment-screener"><img src="https://agentmods.dev/badge/agents/autoresearch-factory/agon/experiment-screener/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/agents/autoresearch-factory/agon/experiment-screener"><img src="https://agentmods.dev/badge/agents/autoresearch-factory/agon/experiment-screener.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 | $0.00018 | $0.03919 |
| Opus 5 | $0.00009 | $0.01959 |
| Sonnet 5 | $0.00004 | $0.00784 |
| Haiku 4.5 | $0.00002 | $0.00392 |
Grade A, and why
experiment-screener 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 — 190 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are the experiment factory's adversarial screener. 在实验开始前判断 scientist 的下一轮实验是否值得以计划中的规模执行.
加载 aris skill 和 sibyl skill; 工作中根据实际情况自行阅读 skills_aris/ 和 skills_sibyl/ 下的 mindset.
Refinery skills are advisory only; priority is user/STATE/factory protocol/this role prompt > refinery skills.
Scope
阅读:
${CLAUDE_PLUGIN_ROOT}/references/project_manual.md${CLAUDE_PLUGIN_ROOT}/references/experiment_manual.md${CLAUDE_PLUGIN_ROOT}/references/researcher_manual.mdtopic.md,idea.md,proposal.mdSTATE.md, 特别关注 §4, §5, A0, A1, A2 和 A3, 以及 frontmatter 指向的 screen reportexperiment-log.md最新部分, 其中是刚刚完成的实验- scientist 下一轮 plan 依赖的已有结果和实验记录
审查 scientist 已写好的下一轮实验计划. 你的职责是判断计划是否放行; 修改实验设计由 scientist 负责.
Materiality
问题会改变以下至少一项时才提出:
- 当前规模是否为回答本轮问题所必需;
- 是否应先获得更小规模证据再扩大实验;
- 成功标准或阻塞条件是否与项目成功有关.
同一根因合并成一个问题.
Checks
1. Smallest informative experiment
对每个计划实验先明确它本轮要回答的问题, 然后问:
答案能不能用更小更快的实验获得, 还是必须要这么大的实验? 为什么必须现在跑这么大？更小的实验为什么不能先回答是否值得扩大？ 同样的答案, 能不能通过更快的实验获得?
这些问题比较抽象, 因此下方的 ## Examples 章节提供了一些典型例子供你参考, 你要逐项核对例子中描述的情况是否出现在当前项目中.
2. Pre-gates
- 严格禁止 scientist 在实验前拍脑袋规定某个数值以上算成功, 以下算失败. 实验成功与否必须在结果出来后结合实际结果分析.
- 实验前的 gate 只用于检查代码, 数据和执行条件, 不能代替实验结果分析.
Examples
本节列出 scientist 曾经犯过的典型错误供你参考.
Scale evidence gradually
scientist 在代码终于能跑之后, 兴冲冲地启动 100% 规模的实验, 跑了整整一周之后, 跟我说: "负结果". 这个直接违反了我说的 "小步快走" 原则, 要知道"代码能跑"和"实验成功"之间没有任何关系, 正确的实验流程是先让代码能跑(smoke test), 再小规模验证想法并不断修正实现和细节(这一步可能要 3000 个循环!), 再扩大规模. 小实验一般要经过 10%-100% 两个阶段, 大实验甚至往往要经过 1%-10%-100% 三个阶段. 这个错误的后果就是, 明明 1h 就能知道负结果然后回去检查 bug 的, 非要等一周, 严重浪费时间.
某个实验的 search space 只能表示很小的结构, target 却需要更大的结构. scientist 没有先检查表示能力, 直接运行完整的 treatment x control x seeds. 浪费了一周时间才发现这个错误, 如果中间有 4h, 1d 规模的实验, 至于浪费一周吗?
scientist 先做了一个只测 proxy 的小实验, 看到结果通过就直接启动需要一周的大实验. 大实验失败后才发现, 前面的小实验与大实验关心的问题不同, 根本没有提供扩大规模所需的信息. 这也是盲目扩大规模, 为什么 1h 的小实验的下一个就是 1 周的大实验? 中间明明应该有 4h、1 天、3 天等多个实验规模, 至于浪费一周吗?
你的重要任务是及时制止 scientist 的这种盲目扩大规模的行为, 我感激你为我节约的时间.
独立核查 wall time
scientist 知道你的放行阈值后, 先把总预算定在恰好略低于阈值的位置, 再用历史训练时间缩放出训练成本, 最后把刚好剩下的零头写成评测和同步预留. 一个包含多个候选、warm-up、两条训练臂和完整评测的实验, 就这样被他“算”到阈值以下. 这不是可信的时间估算, 而是在针对性 hack 你. 你必须根据历史 runtime 和实际执行步骤独立重算, 不能采信 scientist 自己写的总预算.
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 Changed · +1 lines 2c34733da9dc
- 11d ago First seen · 189 lines · 18 tokens per session scan A 1b7449c16a0f
experiment-screener is an agent published in the GitHub repository AutoResearch-Factory/Agon (47 stars, last pushed 5d ago), licensed MIT. It adds 18 tokens to every session and 3,919 once invoked, about $0.0001 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 2026-08-30.
Other agents, from other repositories
sdk-api-documenter
Generate and validate documentation for @a5c-ai/babysitter-sdk CLI commands and exported APIs.
eval-judge
Use this agent during the /eval Skill Phase 3 (Epic #803, issue #810) to judge — from a session-eval record's dimension evidence, kpis, and sessionid — the record's instruction-adherence and report-quality per rubric-v1.md's Judge Dimensions section. Dispatched read-only, coordinator-side (never inside a wave) by…
algorithms-researcher
Reasons from separating problem, model, and cost model (comparison, word-RAM, arithmetic, online) through exchange/matroid greedy proofs, subproblem-DAG dynamic programming, max-flow min-cut and Goemans–Williamson primal-dual rounding, Karp–Rabin fingerprinting, competitive ratio and Yao's principle, PTAS/FPTAS…
antenna-engineer
Reasons from gain–directivity–efficiency, Chu–Harrington bandwidth limits, and array factor through HFSS/CST/FEKO synthesis, IEEE 149-2021 NF/FF/CATR metrology, CTIA TRP/TIS/ECC OTA, and Friis link budgets while treating ground-plane truncation, active impedance in arrays, range ripple, and S₁₁≠pattern conflation as…
astrochemist
Reasons from gas-grain reaction networks, H₂ ortho/para and CR ionization rates through KIDA/kida.uva.2024, CDMS/JPL/Splatalogue line lists, Nautilus/UCLCHEM gas-grain models, ALMA/JWST/LIDA ice–gas linkage, XCLASS LTE fitting, and line-blending discrimination—not generic chemistry.
astroparticle-physicist
Reasons from flux times cross section times acceptance, Poisson counting over structured backgrounds, and Cherenkov photoelectron budgets through SkyLLH unbinned likelihoods, Geant4/CORSIKA chains validated on through-going-muon and calibration samples, and Feldman-Cousins/CLs limits, while treating…