understudy-cli
AGENTS.md1 starsChanged 35 days agoMITView on GitHub
This repository is a fresh implementation with independent history. Treat these rules as release criteria, not suggestions. - Keep this repository on its own new history. Do not graft, merge, or copy history from another repository. - Keep the repository private. The approved review remote is the private understudylabs/understudy-cli repository. Do not make it public, add another remote, publish a package, or make a release without separate approval. - Treat all tracked and packed content as eventually public. Git history…
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 31,774 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 repository8 other files it gives its agents.
- adapt-model-apiSkill
- build-evalsSkill
- check-workloadSkill
- compare-modelsSkill
- recommend-modelsSkill
- rollout-workloadSkill
- setup-understudySkill
- try-modelsSkill
The file
We split this AGENTS.md into the rules it gives an agent
8 rules. About 2 are common, 5 are distinctive, and 1 is very specific to this project.
≈2
Common · estimate
Many files like this have a rule of this kind.
≈5
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
- could be automatic: a setting could enforce it
- Distinctive: line 8
Keep this repository on its own new history.
- Very specific: line 10
Keep the repository private.
- Common: line 13
Treat all tracked and packed content as eventually public.
- Distinctive: line 42
Prefer the established CLI's working behavior and reviewed implementation logic when carrying capabilities forward.
- Distinctive: line 46
When carrying a capability forward:
- Distinctive: line 51
Add focused, wholly synthetic proof.
- Distinctive: line 53
Do not copy history, customer-derived values, fixtures, private comments, or internal-only assumptions.
- Common: line 59
Do not send live traffic or use credentials while source ownership, data origin, or scope is uncertain.
≈ “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
Understudy CLI repository rules
This repository is a fresh implementation with independent history. Treat these rules as release criteria, not suggestions.
Repository trust
- Keep this repository on its own new history. Do not graft, merge, or copy history from another repository.
- Keep the repository private. The approved review remote is the private
understudylabs/understudy-clirepository. Do not make it public, add another remote, publish a package, or make a release without separate approval. - Treat all tracked and packed content as eventually public. Git history is not a secret store.
Content boundary
Tracked files, commits, fixtures, tests, documentation, generated output, and package contents must not contain:
- customer-derived prompts, completions, payloads, traces, or metadata;
- customer, organization, project, workload, request, or user identifiers;
- credentials, tokens, cookies, authorization values, or secret-bearing URLs;
- absolute private filesystem paths or references to private repositories; or
- transformed, redacted, hashed, or summarized versions of customer material.
Fixtures must be wholly invented, minimal, and clearly synthetic. Private
credentials and integration records belong outside every inspected repository
in the per-user global ~/.understudy/ directory. On POSIX systems, private
directories must be owner-only and private files must be readable and writable
only by their owner. Application migration runs belong in that application's
Git-ignored .understudy/, including traces, fixtures, harnesses, results, and
viewers. Keep them out of history and application packages. This does not permit
customer data in this CLI source checkout: use a separate application workspace.
Environment files, logs, and package archives remain excluded from source evidence.
See docs/privacy-and-data-boundaries.md and docs/security.md for the full
boundary.
Reading the previous codebase
Prefer the established CLI's working behavior and reviewed implementation logic when carrying capabilities forward. Adapt that logic to this repository's module structure: thin command handlers, services, transport, private storage, and tests.
When carrying a capability forward:
- State the destination contract in this repository.
- Reuse reviewed, public-safe implementation logic where practical, keeping this repository's architecture and naming and applicable license notices.
- Add focused, wholly synthetic proof.
Do not copy history, customer-derived values, fixtures, private comments, or internal-only assumptions. Inspect reused logic for the content boundary above; do not import whole files or unrelated workflows merely to preserve source shape.
Working boundary
- Do not send live traffic or use credentials while source ownership, data origin, or scope is uncertain.
- Organization-scoped reporting, capture reads, and project, workload, model, and key management are in scope. Match the platform's credential contract: tenant admin APIs accept OAuth or an organization key; customer APIs that require OAuth remain OAuth-only. Explicit workload route changes are thin primitives; do not add automatic workflow judgment.
- Workload migration may capture and reconstruct real interactions and host an agent-authored private harness. Keep transport, recorded-tool matching, limits, and journals deterministic. Workflow, harness adaptations, clusters, reviews, and presentation belong to the maintained Markdown skill and private agent code. Replay mocks never execute customer tools or change routing.
- Monitoring, Desktop, local model serving, and training job execution remain deferred.
- Until automated tracked-and-packed-content checks exist, manual inspection is
evidence only. Do not claim that
.gitignoreor these documents enforce the boundary.
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.

