test-audit

Test Audit

One value bar, three modes:

• Audit (default): a focused sweep for a few high-confidence candidates in one owner area. Land each coherent batch on its own; optimize for confidence, not deletion count.
• Campaign: prune one whole subsystem's test surface (every test file a package, service, or feature area owns). Read CAMPAIGN.md first.
• Authoring gate: judge a new or changed test before it lands.

Orient first

Before judging anything, learn how this repo tests:

• Read root and scoped AGENTS.md, CLAUDE.md, and contributing docs for test rules.
• Find the real test commands in CI config (.github/workflows, .gitlab-ci.yml, etc.), then package.json scripts, Makefile, justfile, mix.exs, pyproject.toml, Cargo.toml, or equivalent. CI is the source of truth.
• Note coverage thresholds, test inventories, snapshot directories, and CI sharding that reference test files by path. Deletions must keep these consistent.
• Record the baseline: run the in-scope tests at a pinned commit and note which fail.

Value bar

A test justifies its maintenance cost by protecting observable behavior, a credible regression, or an independently meaningful contract. A test that must change for a behavior-preserving refactor is suspect, not automatically deletable.

Before judging a candidate, read the complete test and its production owner: entry point, callers, callees, sibling implementations, overlapping tests, and relevant history (git log -L, git blame, the PR or issue that added it). When a test claims dependency-backed behavior, check the dependency source or types directly.

Junk patterns

Audits hunt for these; the authoring gate rejects a new test that matches one.

• assertion-free coverage probes (runs code, asserts nothing meaningful);
• self-comparisons and identity copiers;
• copied fixtures, inventories, manifests, or export lists;
• exact source, import, or string greps;
• snapshots of trivial or unstable output that nobody reviews;
• private-function or call-shape tests duplicated at a real boundary;
• duplicate invocations of the same contract, often copy-pasted with one value changed;
• per-caller replays of a shared helper's tests;
• tests whose only purpose is preserving test-only exports, globals, or wrappers;
• dead production code whose only callers are tests;
• expected values computed by the function under test;
• mocks that implement the asserted behavior, or one mock standing in for different APIs;
• fixtures that supply the result, ordering, or callback the code under test should produce, or persistence asserted against a store the path never writes;
• tests that restate a declared config or flag instead of exercising what it promises;
• negative tests that pass for an unrelated reason, such as a different guard rejecting first or an error path production never reaches;
• names that promise more than the assertions check, such as a "clears the cache" test that asserts the cache was not cleared.