agent-evidence-vectors
AGENTS.md31 starsChanged 9 days agoApache-2.0View on GitHub
Read CONTRIBUTING.md before changing anything; these are the rules an agent most often breaks here. - The pre-push hook runs only the push-time checks, and remote CI is the proof. To run every CI shell step locally first: uv run --with pyyaml python scripts/workflow-steps-gate.py. - A new or moved corpus is a row in corpora/REGISTRY.toml. Never edit the packaging block of pyproject.toml or corpora/README.md; run python3 scripts/corpus-registry.py, which writes both. A corpus id is a published name and never changes.…
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 repositoryOne other file it gives its agents.
- llms.txtllms.txt
The file
We split this AGENTS.md into the rules it gives an agent
8 rules. About 0 are common, 1 is distinctive, and 7 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.
7
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 3
Read CONTRIBUTING.md before changing anything.
- Very specific: line 11
Never write or edit a vector file by hand.
- Very specific: line 13
Never type a count into prose.
- Very specific: line 15
Don't change aee/ to make a vector pass.
- Very specific: line 20
Keep README.md to the first screen.
- Very specific: line 26
Run uv run --frozen --only-group dev ruff check .
- Very specific: line 27
Run actionlint after editing anything under .github/workflows/.
- Very specific: line 30
Never edit pages/ by hand.
≈ “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
Instructions for coding agents
Read CONTRIBUTING.md before changing anything; these are the rules an agent most often breaks here.
- The pre-push hook runs only the push-time checks, and remote CI is the proof. To run every CI
shell step locally first:
uv run --with pyyaml python scripts/workflow-steps-gate.py. - A new or moved corpus is a row in
corpora/REGISTRY.toml. Never edit the packaging block ofpyproject.tomlorcorpora/README.md; runpython3 scripts/corpus-registry.py, which writes both. A corpus id is a published name and never changes. - Never write or edit a vector file by hand. Change the generator in the corpus directory, then
regenerate;
scripts/regenerability-gate.pyrefuses a file no generator produces. - Never type a count into prose. Every published count is derived from a manifest, and
scripts/count-gate.pyrefuses a new count-shaped integer that no source backs. - Don't change
aee/to make a vector pass. If the reference rail must change, that is a separate fix with its own reason. - A question the specification leaves open goes in
vectors/indeterminate/, never into a wider reject code set. - Sign off every commit (
git commit -s); the DCO check refuses a commit without it. - Keep
README.mdto the first screen. Detail goes indocs/, andscripts/readme-lint.pyfails a README over 600 prose words, with no code block in its first 45 lines, or with a relative link that resolves to no file.
Parts imported from agent-evidence (crates/, packages/, actions/, pages/, release.toml):
- Run
uv run --frozen --only-group dev ruff check .anduv run --frozen --only-group dev python -m pytestfrom the root before every commit that touches them. - Run
actionlintafter editing anything under.github/workflows/. - A part adds its CI job to
.github/workflows/parts.ymland to theneedsof itsci-okjob, and its row torelease.toml. - A crate or package release is a published GitHub release whose tag is
<prefix>-v<version>(prefixes inrelease.toml);release.ymlchecks it. A barev<version>tag is the corpus release and publishes only theagent-evidence-vectorswheel. - Never edit
pages/by hand. Rebuild it withscripts/build_predicate_pages.pyand commit the new pin inpages/predicate/v1/PAGES.sha256with the page;docs/predicate-pages.mdsays when a page may change.
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.

