amika / sandbox-image
AGENTS.md301 starsChanged 11 days agoApache-2.0View on GitHub
Source of truth for the coder and coder-plus-docker images. Most of this directory is generated, so an edit here is only half the change. - Never hand-edit generated files. generated/*.Dockerfile, generated/bundle.json, assets/skills/, and everything under go/internal/sandbox/sandbox-image/ are generator output. assets/skills/ in particular is a copy: edit the skill at the repository root (.agents/skills/) and rerun the generator. - Run the generator after any change to manifest.toml, versions.env, steps/, assets/, or verify/, then check it: - Commit the source change and…
- Reads credentials
- 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 repository11 other files it gives its agents.
- AGENTS.mdAGENTS.md
- js/AGENTS.mdAGENTS.md
- amika-annotationsSkill
- amika-cliSkill
- create-cli-release-commitSkill
- create-sdk-release-commitSkill
- distill-sessionSkill
- make-stacked-prsSkill
- naive-reader-reviewSkill
- new-specSkill
- release-commit-sandbox-imageSkill
The file
We split this AGENTS.md into the rules it gives an agent
1 rule. About 0 are common, 0 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 34
Prefer build. A build,boot check pays its cost on every sandbox a customer starts.
≈ “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
Sandbox image — agent guidance
Source of truth for the coder and coder-plus-docker images. Most of this
directory is generated, so an edit here is only half the change.
- Never hand-edit generated files.
generated/*.Dockerfile,generated/bundle.json,assets/skills/, and everything undergo/internal/sandbox/sandbox-image/are generator output.assets/skills/in particular is a copy: edit the skill at the repository root (.agents/skills/) and rerun the generator. Run the generator after any change to
manifest.toml,versions.env,steps/,assets/, orverify/, then check it:python3 sandbox-image/generate.py make test-sandbox-image- Commit the source change and every generated file together. The
go/internal/sandbox/sandbox-image/mirror duplicates this bundle on purpose, sogo installneeds no code-generation step. That duplication is intentional; do not "clean it up". python3 sandbox-image/generate.py --checkverifies generated files are current without writing them.
Verification checks
verify/checks/ is auto-discovered by find (executable *.sh, sorted), so a
new check needs no registration. Each sets CHECK_CONTEXTS:
build— image properties, asserted once at build time.boot— costs every sandbox start. Only for what cannot be checked earlier.
Prefer build. A build,boot check pays its cost on every sandbox a customer
starts.
Paths shared with the control plane
Some paths in this image are claimed by amika-mono at provision time, so a
change here can break something no diff in this repo shows. /usr/local/bin/gh
is one: the app_token gh shim installs there and resolves the real gh by
scanning PATH for one that is not itself, which works only while the image
installs gh elsewhere (16-gh-shim-path-free.sh guards this).
See sandbox-image/README.md for the full change workflow and local builds.
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.

