fix-e2e-failureslisted

# Fix E2E Test Failures This skill handles E2E test failures reported in the ClaudeSpyTestResults repository. It finds the latest failing report, identifies what broke, and guides the fix. ## How failures are reported Screenshot mismatches are **non-fatal** in the orchestrator — after a failed screenshot comparison the scenario keeps running, so a single scenario can produce **multiple failed steps**. Only a non-screenshot error (element not found, timeout, assertion failure, etc.) stops the scenario early. This means each failing scenario may have: - One or more screenshot mismatches (scenario ran to completion), OR - Zero or more screenshot mismatches followed by a fatal functional error (scenario stopped at that point). Handle every failed step — don't stop at the first one. ## Step 1: Find Failures Run the bundled script to pull latest results and extract failures in one step: ```bash ${CLAUDE_SKILL_DIR}/scripts/find_failures.py --results-dir ../ClaudeSpyTestResults ``` **Optional PR-number argument:** If the user invoked the skill with a PR number (e.g. `/fix-e2e-failures 444`), pass it through with `--pr` so the **newest run associated with that PR** is analyzed instead of the most recent failing run across all PRs: ```bash ${CLAUDE_SKILL_DIR}/scripts/find_failures.py --results-dir ../ClaudeSpyTestResults --pr 444 ``` The number matches the `prNumber` field on entries in `ClaudeSpyTestResults/results/index.json`. If no argument is provided, the script default