chore: remove outdated pr-review skill (#3028)

* chore: remove outdated pr-review skill

The .qwen/skills/pr-review skill is outdated and no longer needed.
Also removes the dangling reference from terminal-capture skill.

* chore: add token-stats script to e2e-testing skill

* fix(test): remove flaky AskUserQuestionDialog Submit tab test

The "shows Submit tab for multiple questions" test fails on macOS CI
due to terminal width truncation causing "Submit answers" to render
as just "Submit".

.qwen/skills/e2e-testing/SKILL.md

@@ -156,3 +156,24 @@ tmux kill-session -t test
 For testing MCP tool behavior end-to-end, read `references/mcp-testing.md`. It
 covers the setup gotchas (config location, git repo requirement) and includes
 a reusable zero-dependency test server template in `scripts/mcp-test-server.js`.
+
+## Token Usage Stats
+
+Use `scripts/token-stats.py` to summarize token usage across recent API logs:
+
+```bash
+python3 .qwen/skills/e2e-testing/scripts/token-stats.py 20  # last 20 requests
+```
+
+Shows input, cached, and output tokens per request with cache hit rates. Useful
+for verifying prompt caching behavior or investigating unexpected token counts.
+
+## Tips
+
+- Use interactive (tmux) mode when the bug involves permission prompts, slash
+  commands, or keyboard interactions. Headless mode has no TUI — these don't
+  exist there.
+- Use interactive (tmux) mode for hang-related issues. Headless mode produces
+  no output when the process stalls, giving you nothing to work with.
+- Use `--approval-mode default` when testing permission rules. `yolo` bypasses
+  rule evaluation entirely — it can't test whether a rule matches.

.qwen/skills/e2e-testing/scripts/token-stats.py

@@ -0,0 +1,70 @@
+#!/usr/bin/env python3
+"""Display token usage stats from the last X request logs in ~/.qwen/logs."""
+
+import argparse
+import json
+from pathlib import Path
+
+
+def parse_args():
+    p = argparse.ArgumentParser(description="Show token stats from qwen request logs")
+    p.add_argument("count", nargs="?", type=int, default=10, help="Number of recent logs to show (default: 10)")
+    p.add_argument("--log-dir", default=Path.home() / ".qwen" / "logs", type=Path)
+    return p.parse_args()
+
+
+def load_logs(log_dir: Path, count: int):
+    files = sorted(log_dir.glob("*.json"))
+    for f in files[-count:]:
+        try:
+            with open(f) as fh:
+                yield json.load(fh), f.name
+        except (json.JSONDecodeError, OSError):
+            continue
+
+
+def main():
+    args = parse_args()
+    if not args.log_dir.is_dir():
+        print(f"Log directory not found: {args.log_dir}")
+        return
+
+    rows = []
+    total_input = total_cached = total_output = 0
+
+    for data, fname in load_logs(args.log_dir, args.count):
+        ts = data.get("timestamp", "?")
+        model = data.get("request", {}).get("model", "?")
+        usage = data.get("response", {}).get("usage", {})
+
+        input_tok = usage.get("prompt_tokens", 0)
+        output_tok = usage.get("completion_tokens", 0)
+        cached_tok = usage.get("prompt_tokens_details", {}).get("cached_tokens", 0)
+        cache_rate = (cached_tok / input_tok * 100) if input_tok else 0
+
+        total_input += input_tok
+        total_cached += cached_tok
+        total_output += output_tok
+
+        rows.append((ts, model, input_tok, cached_tok, output_tok, cache_rate))
+
+    if not rows:
+        print("No logs found.")
+        return
+
+    # Print table
+    hdr = f"{'Timestamp':<28} {'Model':<16} {'Input':>8} {'Cached':>8} {'Output':>8} {'Cache%':>7}"
+    sep = "-" * len(hdr)
+    print(hdr)
+    print(sep)
+    for ts, model, inp, cached, out, rate in rows:
+        print(f"{ts:<28} {model:<16} {inp:>8,} {cached:>8,} {out:>8,} {rate:>6.1f}%")
+
+    # Totals
+    print(sep)
+    overall_rate = (total_cached / total_input * 100) if total_input else 0
+    print(f"{'TOTAL':<28} {'':<16} {total_input:>8,} {total_cached:>8,} {total_output:>8,} {overall_rate:>6.1f}%")
+
+
+if __name__ == "__main__":
+    main()