fix(task11-12): simplify review findings — 5 fixes + 1 new test

Reviewers flagged 3 must-fix + 3 minor items on commits 365826d + 7124342.

Important:
- check_prisma_trAIce_freshness.py: catch yaml.YAMLError (previously
  unhandled — malformed YAML would surface as a traceback rather than a
  clean ERROR: line). Added test_malformed_yaml_fails_cleanly covering
  this path (26 tests total, all green).
- Script's upstream_source fallback string was the repo slug
  'cqh4046/PRISMA-trAIce'; the real frontmatter stores the full URL, and
  the fallback should match that shape for user clarity.
- PERFORMANCE.md attribution was imprecise: "ARS does not persist session
  state inside Claude Code" conflates ARS with Claude Code's own session
  mechanism. Rewrite as "ARS does not maintain its own orchestrator state
  between sessions."

Minor:
- Remove dead 'import subprocess' in test file (unused; run_script
  handles subprocess).
- Remove opaque '# Non-blocking: exit 0 per E6 in spec' comment; the
  module docstring already states non-blocking semantics.
- PERFORMANCE.md v3.4.0 cost table: align column headers with the main
  cost table above (Skill / Mode, Input Tokens, Output Tokens, Estimated
  Cost — title-case) and add the "+" prefix to deltas + cross-reference
  the 15K-word / 60-ref estimate basis.

docs/PERFORMANCE.md

@@ -36,16 +36,16 @@
 The full academic pipeline is designed for human-in-the-loop execution, with mandatory user confirmation at every stage. In practice, a full run often spans hours to days — longer than Anthropic's prompt cache TTL (5 minutes). Two consequences:
 
 1. **Cache misses between checkpoints are normal.** When a stage checkpoint pauses longer than 5 minutes, the next stage reads its context uncached. This is an unavoidable cost of human-paced pipelines.
-2. **Cross-session resume relies on Material Passport.** ARS does not persist session state inside Claude Code. To resume in a new session, paste your Material Passport YAML back; the orchestrator reads `compliance_history[]` and stage completion markers to locate your breakpoint.
+2. **Cross-session resume relies on Material Passport.** ARS does not maintain its own orchestrator state between sessions. To resume in a new session, paste your Material Passport YAML back; the orchestrator reads `compliance_history[]` and stage completion markers to locate your breakpoint.
 
 ### v3.4.0 compliance agent cost
 
 Adding the mode-aware `compliance_agent` to Stage 2.5 and Stage 4.5 increases full-pipeline SR tokens by approximately:
 
-| Skill / mode | Input token delta | Output token delta | Estimated cost delta |
+| Skill / Mode | Input Tokens | Output Tokens | Estimated Cost |
 |---|---|---|---|
-| `deep-research systematic-review` (2.5 only) | ~5–8K | ~3–5K | ~$0.15 |
-| Full pipeline SR (2.5 + 4.5) | ~10–15K | ~5–8K | ~$0.30 |
-| `academic-paper full` (pre-finalize) | ~3–5K | ~2–3K | ~$0.08 |
+| `deep-research systematic-review` (2.5 only) | +~5–8K | +~3–5K | +~$0.15 |
+| Full pipeline SR (2.5 + 4.5) | +~10–15K | +~5–8K | +~$0.30 |
+| `academic-paper full` (pre-finalize) | +~3–5K | +~2–3K | +~$0.08 |
 
-These are on top of the existing per-skill costs in the table above. Cross-model verification costs (if enabled) are unchanged.
+These are on top of the existing per-skill costs in the table above (same 15,000-word / 60-reference basis; see footnote on line 23). Cross-model verification costs (if enabled) are unchanged.

docs/PERFORMANCE.zh-TW.md

@@ -36,16 +36,16 @@
 完整 pipeline 設計為 human-in-the-loop，每個階段都需使用者確認。實務上一次完整執行會跨越數小時到數天，遠長於 Anthropic 的 prompt cache TTL（5 分鐘）。兩項結果：
 
 1. **階段間 cache miss 是常態。** 當 stage checkpoint 停留超過 5 分鐘，下一階段會以未快取狀態讀取 context。這是 human-paced pipeline 不可避免的成本。
-2. **跨 session 續跑依賴 Material Passport。** ARS 不在 Claude Code 內保留 session state；要在新 session 續跑，把 Material Passport YAML 貼回即可。orchestrator 讀取 `compliance_history[]` 與階段完成標記定位中斷點。
+2. **跨 session 續跑依賴 Material Passport。** ARS 本身不跨 session 保留 orchestrator 狀態。要在新 session 續跑，把 Material Passport YAML 貼回即可；orchestrator 讀取 `compliance_history[]` 與階段完成標記定位中斷點。
 
 ### v3.4.0 compliance agent 成本
 
 在 Stage 2.5 與 Stage 4.5 加上 mode-aware `compliance_agent` 會讓 SR 全 pipeline token 多出：
 
-| Skill / mode | Input token 增量 | Output token 增量 | 成本增量 |
+| Skill / 模式 | 輸入 Token | 輸出 Token | 估算費用 |
 |---|---|---|---|
-| `deep-research systematic-review`（僅 2.5）| ~5–8K | ~3–5K | ~$0.15 |
-| 全 pipeline SR（2.5 + 4.5）| ~10–15K | ~5–8K | ~$0.30 |
-| `academic-paper full`（pre-finalize）| ~3–5K | ~2–3K | ~$0.08 |
+| `deep-research systematic-review`（僅 2.5）| +~5–8K | +~3–5K | +~$0.15 |
+| 全 pipeline SR（2.5 + 4.5）| +~10–15K | +~5–8K | +~$0.30 |
+| `academic-paper full`（pre-finalize）| +~3–5K | +~2–3K | +~$0.08 |
 
-以上是在既有 per-skill 成本之上額外產生。Cross-model verification 成本（若啟用）維持不變。
+以上為既有 per-skill 成本之上的額外增量（與上表共用 15,000 字 / 60 篇引用基準，見上表下方 footnote）。跨模型驗證成本（若啟用）維持不變。

scripts/check_prisma_trAIce_freshness.py

@@ -62,7 +62,7 @@ def main() -> int:
         text = args.path.read_text(encoding="utf-8")
         fm = extract_frontmatter(text)
         snapshot = parse_snapshot_date(fm)
-    except (FileNotFoundError, ValueError) as exc:
+    except (FileNotFoundError, ValueError, yaml.YAMLError) as exc:
         print(f"ERROR: {exc}", file=sys.stderr)
         return 1
 
@@ -71,11 +71,10 @@ def main() -> int:
         print(
             f"WARNING: prisma_trAIce_protocol.md snapshot is {age_days} days old "
             f"(threshold {STALE_THRESHOLD_DAYS}). Upstream may have updated — "
-            f"please review {fm.get('upstream_source', 'cqh4046/PRISMA-trAIce')} "
+            f"please review {fm.get('upstream_source', 'https://github.com/cqh4046/PRISMA-trAIce')} "
             f"and re-sync if needed. (STALE status surfaced; non-blocking.)",
             file=sys.stderr,
         )
-        # Non-blocking: exit 0 per E6 in spec
     else:
         print(f"OK: snapshot is {age_days} days old (current)")
     return 0

scripts/test_check_prisma_trAIce_freshness.py

@@ -1,5 +1,4 @@
 """Unit tests for check_prisma_trAIce_freshness.py."""
-import subprocess
 import unittest
 from pathlib import Path
 from tempfile import TemporaryDirectory
@@ -53,6 +52,17 @@ class TestFreshnessCheck(unittest.TestCase):
             result = run_script(SCRIPT, str(p))
             self.assertEqual(result.returncode, 1)
 
+    def test_malformed_yaml_fails_cleanly(self) -> None:
+        with TemporaryDirectory() as tmp:
+            p = Path(tmp) / "prisma_trAIce_protocol.md"
+            p.write_text(
+                '---\nsnapshot_date: "2026-03-01\nunclosed_quote: "yes\n---\n# body\n',
+                encoding="utf-8",
+            )
+            result = run_script(SCRIPT, str(p))
+            self.assertEqual(result.returncode, 1)
+            self.assertIn("ERROR", result.stderr)
+
 
 if __name__ == "__main__":
     unittest.main()