Skip to content

AGENTS.md

Latest commit

 

History

History
73 lines (51 loc) · 2.9 KB

AGENTS.md

File metadata and controls

73 lines (51 loc) · 2.9 KB
Raw

AI Agent Instructions

Build & Test

NEVER run pytest, ruff, or mypy directly. Always use bazel commands instead. Direct invocations bypass the build system's environment and dependency setup and produce incorrect results.

Bazel is the primary build system for unit tests, linting, formatting, and type checking. System tests use run_tests.sh:

tests/run_tests.sh test-system-singlesite
tests/run_tests.sh test-system-multisite
tests/run_tests.sh test-system-gui

Always format, lint, and run tests after completing a task. Use bazel run //:format <PATHS> to format and bazel lint --fix to autofix lint findings. Report any remaining non-autofixable findings (including mypy) to the user.

See BAZEL.md for the full developer command reference.

Python Conventions

  • Python version per pyproject.toml; type hints required; pathlib over os.path
  • Formatting/linting via bazel lint (ruff); type checking via bazel build --config=mypy (mypy)
  • Agent plugins: Python 3.4+ compatible
  • GUI cannot import cmk/base internals (enforced by component isolation)

Testing

Read TESTING.md before writing, changing, or debugging tests, and run its self-check before finishing. It defines the test levels and their directories, the flaky-test process, and the core rules: test through the public surface, inject dependencies instead of patching Checkmk code, one behavior per test.

Editions

The codebase supports five editions: community, pro, ultimate, ultimatemt, cloud. The active edition controls which cmk targets and Python modules are available; pass --cmk_edition=<edition> to Bazel.

Skills & Agents

Reusable AI workflows for this repo (code review, werk authoring, Gerrit/Jenkins/Jira/crash-report helpers, backports, etc.) are maintained as Claude Code plugins in the internal marketplace repository: https://github.com/Checkmk/checkmk-claude-marketplace

Notable entry points:

  • code-review agent — after committing a change (or a relation chain of commits), consider running it, or propose doing so, to catch logical, semantic, and architectural issues before pushing for review.
  • /test-review skill — run it whenever implementing or changing tests.

Do not add skills or agents under .claude/ or .github/ in this repo — contribute them to the marketplace instead.

Commit rules

Before a commit do a sanity check of your changes and run the linters and formatters.

Files that must never be committed

Ensure the following paths are NOT part of any commit (check git status / git diff --cached before committing and unstage them if they appear):

  • tests/qa-test-data (submodule — never update its pinned commit)