triage-red-suitelisted

# Skill: triage-red-suite Take a suite with many failures and return an honest account of it. The instruction "make it green" and the goal "make it trustworthy" come apart precisely here, and the gap is where suites go to die. Every red test is one of two things: a defect in the test, or a message about the product. Fixing the first is the job. Silencing the second destroys the reason the suite exists — and it is invisible afterwards, because a suppressed signal looks exactly like no signal. Nine of ten fixes to a red suite are legitimate. It is the tenth that determines whether anyone trusts the green. ## Preconditions `.claude/test-profile.md` exists — you need the run commands and the repo's conventions before changing test code. For one failing test, dispatch **playwright-test-healer** (browser) or **playwright-api-test-healer** (request) instead; this skill is for volume. ## Procedure ### 1. Get a clean, complete failure list Run the suite with retries off and a line-oriented reporter. Retries on will hide flakes as passes and you will fix the wrong things. Capture every failure with its actual error. Do not start fixing while the run is in progress — triage-before-fix is what stops you from patching six symptoms of one cause. ### 2. Look for the single cause first Before classifying individually, check whether the failures share one root. Common shapes: - **All of them, immediately** — environment down, wrong target, missing credentials, expired auth. Fix