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 jdpalumbo2/workbench --skill empirical-falsificationgit clone --depth 1 https://github.com/jdpalumbo2/workbenchWrote 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/jdpalumbo2/workbench/empirical-falsification)<a href="https://agentmods.dev/skills/jdpalumbo2/workbench/empirical-falsification"><img src="https://agentmods.dev/badge/skills/jdpalumbo2/workbench/empirical-falsification/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/jdpalumbo2/workbench/empirical-falsification"><img src="https://agentmods.dev/badge/skills/jdpalumbo2/workbench/empirical-falsification.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 | $0.00046 | $0.00635 |
| Opus 5 | $0.00023 | $0.00318 |
| Sonnet 5 | $0.00009 | $0.00127 |
| Haiku 4.5 | $0.00005 | $0.00064 |
Grade A, and why
empirical-falsification 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 9d 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 — 40 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Empirical Falsification
Overview
Check a claim against actual evidence — pass or fail — instead of against how plausible it sounds. Not vibes, not logic, not "this matches what I already believe": evidence.
The most expensive thing in any strategy is a claim nobody checked.
When to Use
- A statistic or claim is about to be repeated in a deck, meeting, or document.
- You're building on an assumption that's never actually been verified.
- A "fact everyone knows" starts to feel shaky when you try to source it.
- Stakes are high and you don't want to be the one challenged on it in the room.
Core Pattern
- State the claim precisely. Vague claims can't be falsified — pin down the exact assertion, including any numbers or scope ("70% of X leave because of Y," not "people leave because of managers").
- Search for real evidence, not recalled evidence. This is the step that matters: do not answer from training-data memory of what sources probably say. Use WebSearch (and WebFetch to check a source in full) to find what current, citable sources actually say.
- Report pass or fail against what was found, with sources. If evidence is mixed or the claim is more nuanced than stated, say so — "partially true, with this caveat" is a valid, useful result, not a failure to reach a verdict.
Implementation
This skill requires the WebSearch tool. Do not produce citations from memory — an LLM will confidently generate plausible-sounding sources that don't check out, which is the exact failure mode this skill exists to prevent. If WebSearch is unavailable, say so explicitly rather than falling back to recalled citations.
When a citation leads to another citation instead of a primary source, follow the chain — a stat repeated everywhere with nothing at the bottom of the chain is itself the finding. Stop once you reach either a primary source (the actual dataset, study, or report) or two consecutive dead ends (a broken link, a page that doesn't contain the number it's cited for). Report which one you hit.
What ships with it
1 file 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.
- 9d ago First seen · 40 lines · 46 tokens per session scan A 1a4fc3dab4e4
empirical-falsification is a skill published in the GitHub repository jdpalumbo2/workbench (1 stars, last pushed 6d ago), licensed MIT. It adds 46 tokens to every session and 635 once invoked, about $0.0002 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 skills, from other repositories
code-review-excellence
This skill should be used when the user asks to review a diff or pull request, write review comments, audit code quality, establish review standards, or improve how a team performs code review.
git-workflow
This skill should be used when the user asks to "create git commit", "manage branches", "follow git workflow", "use Conventional Commits", "handle merge conflicts", or asks about git branching strategies, version control best practices, pull request workflows. Provides comprehensive Git workflow guidance for team…
architecture-design
Use only when creating new registrable ML components that require Factory or Registry patterns.
results-analysis
This skill should be used when the user asks to "analyze experimental results", "run strict statistical analysis", "compare model performance", "generate scientific figures", "check significance", "do ablation analysis", or mentions interpreting experiment data with rigorous statistics and visualization. It focuses on…
review-response
Systematic review response workflow from comment analysis to professional rebuttal writing. Use when the user asks to "write rebuttal", "respond to reviewers", "draft review response", or "analyze review comments". Improves paper acceptance rates.
daily-paper-generator
Use when the user asks to generate daily paper digests on a general topic. This skill supports both arXiv and bioRxiv (or either one), then produces structured Chinese/English summaries for selected papers.