test-audit

User Input

$ARGUMENTS

$ARGUMENTS can contain:

• A path scope (e.g. tests/auth or src/) -- limits the audit to that subtree. Default: the whole repo.
• --summary -- print the report to the terminal only, do not write a file.

Setup

1. Read .coco/config.yaml for:

   • testing.test_audit_report_dir (default: docs/test-audit)
   • testing.test_audit_exclude -- path globs to skip
   • testing.test_command -- context only; this command never runs the suite
   • project.specs_dir (default: specs)

2. Load the test value rubric, resolving in order:

   • .coco/templates/test-value-rubric.md (project override)
   • ${CLAUDE_PLUGIN_ROOT}/templates/test-value-rubric.md (default)

3. Get the current date for the report filename:

   date -u +%Y-%m-%d
   

Execution

1. Inventory the suite

Find test files by convention, honoring the scope argument and test_audit_exclude:

git ls-files

Filter the output for test paths -- test_*, *_test.*, *.test.*, *.spec.*, tests/**, spec/**, __tests__/** -- and drop anything matching an exclude glob. Use git ls-files rather than find so untracked scratch files and build artifacts never enter the audit.

Report the count before proceeding. If it is zero, follow Edge Cases.

2. Load the stated intent

For each feature directory under {specs_dir}/, read the ## Test Strategy section of design.md if present. Collect the FRs marked Test? = yes, the levels chosen, and everything listed under Not worth testing.

This is evidence, not authority. Designs are written before implementation and are never rewritten afterward, so a stale "not worth testing" note penalizes a test's behavioral anchoring but never by itself justifies deleting it. Record which designs were found -- the report states what the audit had to work with.

3. Score

If the inventory is 40 files or fewer: score inline, reading each test file and enough of the code under test to answer "what defect would ship without this?"