shadowclone / evals
AGENTS.md25 starsChanged 32 days agoMITView on GitHub
Private evaluation suites. They measure whether agents follow preferences and work through pull requests. Maintainers run them; the CLI does not. The code review benchmark is the separate code-review-bench repository. (generated by bun run guides; do not edit by hand) Run bun test in this folder, or bun test evals from the root. The tests use synthetic fixtures and mock runners, so they make no paid call. Some tests run git. Update this file in the same commit when you…
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 31,700 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 repository49 other files it gives its agents.
- AGENTS.mdAGENTS.md
- tooling/AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- evals/CLAUDE.mdCLAUDE.md
- tooling/CLAUDE.mdCLAUDE.md
- choose-by-consequenceSkill
- design-deep-modulesSkill
- diagnose-before-editingSkill
- plan-with-review-pageSkill
- research-primary-sourcesSkill
- resolve-conflicts-by-intentSkill
- scope-confirmed-changesSkill
- shadowclone-build-preferencesSkill
- shadowclone-reviewSkill
- shadowclone-workSkill
- tests-that-catch-bugsSkill
- typescript-type-safetySkill
- verify-and-reviewSkill
- write-plain-englishSkill
- choose-by-consequenceSkill
- data-handlingSkill
- design-deep-modulesSkill
- diagnose-before-editingSkill
- plan-with-review-pageSkill
- research-primary-sourcesSkill
- resolve-conflicts-by-intentSkill
- scope-confirmed-changesSkill
- shadowclone-build-preferencesSkill
- shadowclone-reviewSkill
- shadowclone-workSkill
- tests-that-catch-bugsSkill
- typescript-type-safetySkill
- verify-and-reviewSkill
- write-plain-englishSkill
- choose-by-consequenceSkill
- design-deep-modulesSkill
- diagnose-before-editingSkill
- guided-pr-reviewSkill
- plan-with-review-pageSkill
- research-primary-sourcesSkill
- resolve-conflicts-by-intentSkill
- scope-confirmed-changesSkill
- shadowclone-reviewSkill
- shadowclone-workSkill
- tests-that-catch-bugsSkill
- typescript-type-safetySkill
- verify-and-reviewSkill
- verify-review-findingsSkill
- write-plain-englishSkill
The file
We split this AGENTS.md into the rules it gives an agent
3 rules. About 0 are common, 1 is distinctive, and 2 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.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 31
Do not change a number in evals.md or the README results table without a new run.
- Very specific: line 35
Run bun test in this folder, or bun test evals from the root.
- Distinctive: line 37
Update this file in the same commit when you change the package purpose, public exports, dependencies, or a rule above.
≈ “Common” is an estimate: at least 1 in 5 of the 40,246 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
@shadowclone/evals
Private evaluation suites. They measure whether agents follow preferences and work through pull requests. Maintainers run them; the CLI does not. The code review benchmark is the separate code-review-bench repository.
<!-- package-facts:start -->(generated by bun run guides; do not edit by hand)
- Package name:
@shadowclone/evals - Workspace dependencies:
@shadowclone/agents,@shadowclone/cli,@shadowclone/core,@shadowclone/environment,@shadowclone/learning,@shadowclone/redact,@shadowclone/sessions,@shadowclone/skills - External dependencies:
commander,typescript,zod - Export subpaths: none
<!-- package-facts:end -->Owns
run.ts: thebun run evalentry.cli/holds one handler for each protocol.fixed/: the frozen preference benchmark (preference-respect-v3), its cases, graders, and held-out manifest.native/: native agent runs, private storage checks, and receipts.work/: theshadowclone-worksuite forbun run eval:work.no-comments/: the comment suite.shared/: ledgers, locks, and isolation.
Rules
- A phase that calls a model needs an
approve-*step,--yes, and a scope that names the host, model, and call ceiling. - Private storage cannot be inside a repository.
requirePrivateDirectoryenforces this. - Held-out cases stay outside the checkout. Only their hashes and family names are public.
- Frozen guidance includes generated task catalogs and their native lookup instruction. Catalog paths remain portable with the other guidance files.
- A report stays incomplete while a reserved call has no recorded attempt.
- Do not change a number in
evals.mdor the README results table without a new run.
Tests
Run bun test in this folder, or bun test evals from the root. The tests use synthetic fixtures and mock runners, so they make no paid call. Some tests run git.
Update this file in the same commit when you change the package purpose, public exports, dependencies, or a rule above.
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.