agentleFS
Sign inStart free
AWSKnown publisherawslabs/ml-container-creator/CLAUDE.md

ml-container-creator

CLAUDE.md20 starsChanged 17 days agoApache-2.0View on GitHub

This repo is also worked on by Kiro, which may be running at the same time. The shared rules live in the Kiro steering files below and are imported, not copied, so the two tools cannot drift. Where an import disagrees with this file, this file wins (the imports have a few stale details, listed under Corrections). @.kiro/steering.md @.kiro/steering/derive-dont-hardcode.md

  • Commits and pushes

The file

We split this CLAUDE.md into the rules it gives an agent

11 rules. About 0 are common, 3 are distinctive, and 8 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 8

    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
  1. Very specific:

    Treat .kiro/ as read-only: Never edit, move or delete anything under .kiro/ (steering, specs, trackers, backlog.jsonl, tools/).

    line 9
  2. Very specific:

    Read specs and trackers freely: .kiro/specs/<name>/ (requirements, design, tasks), .kiro/backlog.jsonl, .kiro/v2.0-tracker.md.

    line 10
  3. Very specific:

    Only .kiro/steering.md and .kiro/hyperpod-ops-surface.md are tracked in git.

    line 11
  4. Very specific:

    Do not create AGENTS.md, .kiro/steering/* or anything else Kiro would load automatically.

    line 12
  5. Distinctive:

    Run git status before editing.

    line 13
  6. Very specific:

    Never run npm run codegen, tests or git operations that rewrite many files while Kiro may be doing the same.

    line 14
  7. Very specific:

    Do not infer implementation status from the package.json version.

    line 20
  8. Very specific:

    Use plan mode for any BL item.

    line 30
  9. Distinctive:

    When a test fails, decide whether the test or the code is wrong using the spec, committed behavior and git history.

    line 31
  10. Distinctive:

    Never fabricate data to make a test pass.

    line 32
  11. Very specific:

    Ask before git commit, git push, npm publish, npx cdk, aws and node scripts/e2e-runner.js.

    line 37

≈ “Common” is an estimate: at least 1 in 5 of the 26,980 CLAUDE.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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
CLAUDE.md

ml-container-creator (MLCC) — Claude Code notes

This repo is also worked on by Kiro, which may be running at the same time. The shared rules live in the Kiro steering files below and are imported, not copied, so the two tools cannot drift. Where an import disagrees with this file, this file wins (the imports have a few stale details, listed under Corrections).

@.kiro/steering.md @.kiro/steering/derive-dont-hardcode.md

Kiro coexistence (read first)

  • Treat .kiro/ as read-only. Never edit, move or delete anything under .kiro/ (steering, specs, trackers, backlog.jsonl, tools/). Kiro owns it and may be mid-edit. If a tracker or spec needs an update, say so and let the user make it.
  • Read specs and trackers freely: .kiro/specs/<name>/ (requirements, design, tasks), .kiro/backlog.jsonl, .kiro/v2.0-tracker.md.
  • Only .kiro/steering.md and .kiro/hyperpod-ops-surface.md are tracked in git. The rest of .kiro/ is local-only (gitignored), so it exists only in this checkout.
  • Do not create AGENTS.md, .kiro/steering/* or anything else Kiro would load automatically.
  • Run git status before editing. If files outside your task are already modified, Kiro is probably working: leave them alone and avoid broad git add -A; stage files by name.
  • Never run npm run codegen, tests or git operations that rewrite many files while Kiro may be doing the same. Ask the user if unsure.

Corrections to the imported steering

  • Regenerate code with npm run codegen (there is no npm run generate).
  • The commit gate is .husky/pre-commit, not just ESLint. In order: reject hardcoded numRuns in staged property tests; lint-staged; Python ast.parse on staged .py (skips EJS templates containing <%); codegen parity when schema, codegen scripts or catalogs are staged (plus catalog and tune-manifest validation); node scripts/sync-command-generator.js --check; npm run build in infra/bootstrap-modules when .ts files are staged; strip __pycache__. Never bypass it with --no-verify. .git/config points core.hooksPath at .husky/_, so .githooks/pre-commit is a legacy duplicate and not the live gate.
  • There are 13 bundled MCP servers in servers/ (the steering text says 12 and lists 11). They are MLCC's product, not dev tooling.
  • Do not infer implementation status from the package.json version. Check the tracker and the code. Specs for a release often sit at drafted or hardened long after the minor bump.

Extra rules not in the steering files

  • Flexible Training Plan (trainingPlanArn) is NOT a bootstrap-time input. Do not add it as a bootstrap option, SSM value or CDK context. The integration point is undecided (likely do/deploy --target hyperpod-eks or a do/config variable). Ask before implementing.
  • MLflow registered model names must not contain / (breaks loading, MLflow #8801). Use -- between family/variant and __ between sub-components.
  • do/ scripts that call python3 must call _require_python_env (from lib/script-contract.sh) after sourcing script-contract.sh, so the project venv (.mlcc/hey-venv) auto-activates.
  • In deploy.d/hyperpod-eks, anything exported inside $(...) does not reach the parent shell. Set HP_MODEL_SOURCE explicitly before envsubst.
  • Serving on HyperPod is vLLM via the inference operator. Do not add SGLang references.

Working from specs (Claude Code)

  • Use plan mode for any BL item. Read its spec and its row in .kiro/v2.0-tracker.md first, then work tasks.md in order and report which tasks you finished.
  • When a test fails, decide whether the test or the code is wrong using the spec, committed behavior and git history. Do not fix whichever side is easier.
  • Never fabricate data to make a test pass. Use a documented skip.

Claude Code guardrails

  • .claude/settings.json denies edits to src/lib/generated/**, and a PreToolUse hook (.claude/hooks/guard-generated.mjs) blocks edits inside # >>> GENERATED ... # <<< END GENERATED blocks. The hook only covers Edit/Write, so never use sed -i or shell redirection on generated files either. Edit the schema or descriptor and run npm run codegen.
  • .mcp.json is Claude-only and loads just e2e-status and agent-knowledge. Kiro has no MCP config in this setup.
  • Ask before git commit, git push, npm publish, npx cdk, aws and node scripts/e2e-runner.js.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.

Check your setup, free