feat(skill): Add flaky-test-detection skill

domains/coding/skills/flaky-test-detection/references/gh-analysis.md

@@ -0,0 +1,113 @@
+# Flaky Test Audit — GitHub Actions
+
+## Prerequisites
+
+```bash
+gh auth status
+git remote get-url origin  # must contain MetaMask/metamask-mobile
+```
+
+## Step 1 — Find the exact unit-test workflow name
+
+```bash
+# List recent runs across all workflows — note the name column
+gh run list --repo MetaMask/metamask-mobile --limit 50
+
+# Confirm the correct name by filtering (adjust if different from "Unit Tests")
+gh run list --repo MetaMask/metamask-mobile --workflow "Unit Tests" --limit 5
+```
+
+Use the **exact** workflow name returned by `gh run list` (e.g. `"Unit Tests"`, `"Jest"`, `"CI"`). Substitute it in every `--workflow` flag in the steps below. If the filter returns 0 results, the name is wrong — re-check Step 1.
+
+## Step 2 — Collect IDs of failed runs
+
+Replace `"Unit Tests"` with the name found in Step 1.
+
+```bash
+FAILED_RUN_IDS=$(gh run list \
+  --repo MetaMask/metamask-mobile \
+  --workflow "Unit Tests" \
+  --limit 200 \
+  --json databaseId,conclusion \
+  --jq '.[] | select(.conclusion == "failure") | .databaseId')
+```
+
+## Step 3 — Extract failing test file names
+
+```bash
+gh run view <run-id> --repo MetaMask/metamask-mobile --log-failed \
+  | grep -oE 'FAIL .+\.test\.(ts|tsx|js|jsx)' \
+  | sed 's/^FAIL //'
+```
+
+## Step 4 — Compute per-file failure count and rate
+
+`TOTAL_RUNS` is all runs in the sampled window (not just failures) — the denominator for rate calculation. It is fetched separately at the end of the script.
+
+> Requires Bash 4+ (for `declare -A`). macOS ships Bash 3.2 by default — run this with `brew install bash` or via `gh`'s bundled environment, or invoke as `bash script.sh` with a Bash 4+ binary on `PATH`.
+
+```bash
+declare -A fail_count
+
+for run_id in $FAILED_RUN_IDS; do
+  files=$(gh run view "$run_id" --repo MetaMask/metamask-mobile --log-failed 2>/dev/null \
+    | grep -oE 'FAIL .+\.test\.(ts|tsx|js|jsx)' \
+    | sed 's/^FAIL //')
+  for f in $files; do
+    fail_count[$f]=$(( ${fail_count[$f]:-0} + 1 ))
+  done
+done
+
+# Count total runs sampled (all runs in the window, not just failures)
+TOTAL_RUNS=$(gh run list \
+  --repo MetaMask/metamask-mobile \
+  --workflow "Unit Tests" \
+  --limit 200 \
+  --json databaseId \
+  --jq 'length')
+
+# Print: failures total rate% file
+for f in "${!fail_count[@]}"; do
+  count=${fail_count[$f]}
+  rate=$(echo "scale=2; $count * 100 / $TOTAL_RUNS" | bc)
+  echo "$count $TOTAL_RUNS ${rate}% $f"
+done | sort -rn
+```
+
+> Rate limit: ~5,000 gh API requests/hour. Add `sleep 1` between loop iterations for large histories (>100 runs).
+
+## Step 5 — Ranked output format
+
+Columns: rank, test file path, failure count, total runs sampled, failure rate, pattern category (from symptom mapping below).
+
+```
+Rank | Test file                                             | Failures | Runs | Rate  | Category
+-----|-------------------------------------------------------|---------:|-----:|------:|---------
+  1  | app/components/Views/BankDetails/BankDetails.test.tsx |      12  |  50  | 24 %  | J1
+  2  | app/util/transactions/transactions.test.ts            |       9  |  50  | 18 %  | J3
+```
+
+Assign the category by matching the failure log symptoms in the table below, then open the pattern section in the skill for the fix.
+
+## Symptom → category mapping
+
+| Symptom in failed log | Category |
+|---|---|
+| `TypeError: terminated` / `SocketError: other side closed` | J1 |
+| Timer / timeout mismatch | J2 |
+| `toHaveBeenCalledTimes` wrong count | J3 |
+| `waitFor` never resolved | J4 or J8 |
+| `Cannot read property of undefined` from store selector | J5 |
+| Test timed out (5 s default) | J6 |
+| Assertion on `Date` or random value fails | J7 |
+| Test hangs indefinitely under fake timers | J8 |
+| Test passes in isolation but fails in suite | J9 |
+| Unexpected function called / wrong return value | J10 |
+
+## Thresholds
+
+| Rate | Action |
+|------|--------|
+| ≥ 10 % | Actively flaky — fix before merging new tests in that file |
+| 5–10 % | At-risk — review for patterns in its category |
+| < 5 % | Monitor |