gallop
AGENTS.md1 starsChanged 44 days agoMITView on GitHub
gallop is a theoretical framework: eight agent skills in markdown and the docs site that presents them. The repository holds no package, plugin, server or eval harness. Six positions: routing, the measurement floor, three method buckets (description, causation, prediction), and the knowledge ceiling. Every skill traces to exactly one; causation carries three skills, description and prediction one each. A proposal that needs a seventh position is refused, not accommodated. - Gerund-form kebab-case directory names, matching the name field exactly. -…
- Installs packages
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 32,671 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository9 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- building-modelsSkill
- choosing-causal-designsSkill
- defining-metricsSkill
- designing-experimentsSkill
- forming-hypothesesSkill
- reading-experimentsSkill
- routing-questionsSkill
- writing-reportsSkill
The file
We split this AGENTS.md into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 41
No text, label, registry line or task id from a holdout enters the repository, a pull request or a note.
≈ “Common” is an estimate: at least 1 in 5 of the 41,581 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Repo conventions
gallop is a theoretical framework: eight agent skills in markdown and the docs site that presents them. The repository holds no package, plugin, server or eval harness.
The taxonomy is the gate
Six positions: routing, the measurement floor, three method buckets (description, causation, prediction), and the knowledge ceiling. Every skill traces to exactly one; causation carries three skills, description and prediction one each. A proposal that needs a seventh position is refused, not accommodated.
Skills
- Gerund-form kebab-case directory names, matching the
namefield exactly. SKILL.mdbody under 500 lines. If it will not fit, cut coverage. Never split a skill to get under the limit; one skill per position.- References sit exactly one level below
SKILL.md. Never deeper: Claude partially reads files reached through a chain. - Reference files are named for their content,
interference.md, notadvanced.md. - Forward slashes everywhere.
- A skill states the method, never a command: the test, the threshold, the inputs and what to report.
Site
site/build.py renders the skill pages, the knowledge, install and about pages,
and the README skills table from the skills and site/content/. Run it after
changing a skill, and python3 site/build.py --check before committing. Its
one dependency is pinned in requirements.txt: pip install -r requirements.txt.
CI runs the same check on every pull request.
Assets
Nothing whose provenance is unclear enters the history. Git history is permanent and deleting a file later does not remove it. No font binaries, no inlined font data, no icons of borrowed lineage.
No text, label, registry line or task id from a holdout enters the repository, a pull request or a note.
Prose
Flat and declarative. State the finding, give the evidence, give the fix. No hype adjectives, no em dashes, no emoji in code or commit messages.
Running the skills
docs/orchestration.md is the wiring: the path a question takes through the
skills and the state they share.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

