Ha Quality Scale Verify

Verify Quality Scale Rule

You are verifying whether a Home Assistant integration follows a quality scale rule. Verify one rule at a time; to check a full tier, verify each of that tier's rules (parallel subagents when possible).

Reading the code is not enough. Many rules have an objective threshold or a validator that CI enforces — run the tool and read its result instead of inferring from the source. Report the measured number or the tool's verdict; when a rule is a judgment from reading code, say so; when you could not check it, mark it unverified rather than implying a pass.

1. Understand the rule

Fetch the rule doc from https://raw.githubusercontent.com/home-assistant/developers.home-assistant/refs/heads/master/docs/core/integration-quality-scale/rules/{rule_name}.md and read the exact requirement — the wording matters (e.g. test-coverage is "Above 95% test coverage for all integration modules": per module, strictly above 95%). Note the tier the rule belongs to, its required patterns, and its exemption criteria.

2. Inspect the integration

Verify the code as it currently is (local working tree, or a checked-out PR head), not the base branch. In homeassistant/components/<domain>, read manifest.json (declared quality_scale tier), quality_scale.yaml (each rule's done/todo/exempt), and the modules and tests/components/<domain> relevant to the rule.

Other sources: the library on PyPI (https://pypi.org/pypi/<package>/json); and the integration docs, resolved in this order — the docs change under review, its linked docs PR, the next branch, then the current branch of home-assistant.io (.../home-assistant.io/refs/heads/<branch>/source/_integrations/<domain>.markdown). Docs for a tier bump usually land on next or a docs PR, so current alone is often stale.

3. Which rules to verify

• Tier bump (the quality_scale value in manifest.json changes): verify every rule from Bronze up to the target tier, from scratch — do not trust the existing done/exempt marks.
• Rule status changes, same tier (flips in quality_scale.yaml, or a new scorecard): verify every rule whose status the diff changes, in any direction. A revert to todo at or below the declared tier breaks the tier claim.
• Code change only (no quality_scale.yaml or tier change): verify the changed code still satisfies the rules in force at the integration's current tier. Map the diff to the rules it implicates and check those — the bar is no regression.
• A single requested rule: verify just that rule.

4. How to prove each rule

Run checks in the project's dev environment. Match the versions the change pins (ruff from requirements_test_pre_commit.txt, syrupy from requirements_test.txt, the library from manifest.json) — version drift gives false passes and false failures. A missing or stale dependency (ModuleNotFoundError, mypy flagging library symbols) is a build-environment gap to fix, not a rule violation; don't assume an unfamiliar imported package is a mistake without confirming it is absent from the project's requirements.

• hassfest (python -m script.hassfest --integration-path homeassistant/components/<domain>) — validates quality_scale.yaml and the validator-backed rules: config-flow, runtime-data, test-before-setup, unique-config-entry, discovery, reconfiguration-flow, strict-typing wiring. Caveats: a rule's validator runs only when that rule is done, so a clean run does not prove a todo/exempt rule — check those directly. --integration-path runs only the per-integration plugins, so the repo-wide ones do not run — notably mypy_config (needed for strict-typing, below) and the brand-metadata check that the config_flow plugin performs only on a full run.
• strict-typing (platinum) — run mypy on the integration. It must be listed in .strict-typing (else mypy runs non-strict and passes vacuously), and mypy.ini must be in sync — validate that (python3 -m script.hassfest -p mypy_config --action validate); never let hassfest regenerate it (the default, which hides staleness). mypy_config is repo-wide, so scope a failure to the target before reporting it.
• test-coverage (silver+) — regenerate translations first (tests load the generated translations/en.json, not strings.json), then run coverage. The rule is strictly above 95% per module; report every module not above 95% by exact covered/total (a rounded "95%" can be 94.7%), with missing lines.
• config-flow-test-coverage (bronze) — config_flow.py fully covered, and every supported flow tested through error recovery to its correct terminal result: CREATE_ENTRY for user/discovery/import setup, ABORT + reauth_successful/reconfigure_successful for reauth/reconfigure, plus proof that a duplicate or otherwise-disallowed second entry is rejected — require the test to assert that the flow aborts, but accept the integration's reason (e.g. already_configured, single_instance_allowed, already_configured_as_subentry). Options flows count and live in config_flow.py, so they are already in scope. 100% line coverage without these scenarios still fails.
• docs-* — confirm the required sections exist in the docs (resolved as in §2). A section absent from readable sources is a finding; mark unverified only when the docs could not be reached.
• brands (bronze) — hassfest does not check this rule. A full run validates brand metadata (the in-repo homeassistant/brands/*.json) via the config_flow plugin, but never the icon/logo image assets, which live in the separate home-assistant/brands repo. Verify the required assets there or in the linked brands PR, else mark unverified.
• Everything else — verify by reading the code against the rule doc.

Rules are cumulative and enforced at or above their own tier (e.g. test-coverage is Silver — it does not fail a Bronze claim). A done rule whose check fails is a finding; so is an exempt reason that masks an implementation gap rather than genuine non-applicability.

5. Report findings

Report the rules with issues and the rules you could not check. For each issue: the rule and the problem (non-compliance, over-claimed done, or invalid exempt), the evidence (measured result / tool output / file location), and the fix. For each unchecked rule: mark it unverified and say what is needed. State which checks you ran. If every rule you could check passes and nothing is left unverified, say so in one line.