[codex] Add report reproducibility metadata

README.md

@@ -142,7 +142,7 @@ Community evidence is shape-checked by GitHub Actions, but publication still req
 | Version | `v2.3` |
 | Audit steps | 14 |
 | Risk matrix | 6D |
-| pytest collected tests | 778 |
+| pytest collected tests | 782 |
 | CLI flags | 21 |
 | Runtime profiles | `general`, `web3`, `full` |
 
@@ -328,7 +328,7 @@ API Relay Audit 也可以作为 agent skill 使用。
 | 版本 | `v2.3` |
 | 审计步骤 | 14 |
 | 风险矩阵 | 6D |
-| pytest collected tests | 778 |
+| pytest collected tests | 782 |
 | CLI flags | 21 |
 | Runtime profiles | `general`, `web3`, `full` |
 

ROADMAP.md

@@ -40,10 +40,11 @@ contributor, arXiv:2026-04-26, 正交威胁轴：模型替换质量欺诈 vs 我
   detector, not a hosted preflight service, not a new API family, and not a
   change to LOW/MEDIUM/HIGH semantics. `inconclusive` remains
   `inconclusive`.
-- **Final test count**: 778/778 passing (733 baseline → 764 after current
+- **Final test count**: 782/782 passing (733 baseline → 764 after current
   master follow-ups → 769 after release-engineering version sync coverage →
   775 after query-family growth contracts → 778 after release-hardening
-  public verification regressions).
+  public verification regressions → 779 after report reproducibility metadata
+  → 782 after metadata source-boundary regressions).
 
 ### v1.9 — Dual-distribution full generation (2026-06-01)
 - **Standalone product promise preserved**: root `audit.py` stays committed,

api_relay_audit/reporter.py

@@ -1,6 +1,6 @@
 """Markdown report generator for audit results."""
 
-from datetime import datetime
+from datetime import datetime, timezone
 
 
 class Reporter:
@@ -77,7 +77,8 @@ def flag(self, level, msg):
         self.summary.append((level, msg))
         self.sections.append(f"{icon} **{msg}**\n")
 
-    def render(self, target_url="", model=""):
+    def render(self, target_url="", model="", tool_version="", profile="",
+               tool_commit=""):
         """Render the complete Markdown report.
 
         Produces a header block (title, metadata, risk summary) followed
@@ -88,6 +89,11 @@ def render(self, target_url="", model=""):
                 metadata when provided.
             model: The model identifier used for the audit. Shown in the
                 report metadata when provided.
+            tool_version: API Relay Audit version used for the run.
+            profile: Audit profile used for the run (``general``, ``web3``,
+                or ``full``).
+            tool_commit: Optional git commit for checkout-based runs. Omitted
+                when the standalone script is run outside a repository.
 
         Returns:
             A single Markdown string containing the full report.
@@ -100,12 +106,18 @@ def render(self, target_url="", model=""):
         """
         header = (
             f"# API Relay Security Audit Report\n\n"
-            f"**Generated**: {datetime.now().strftime('%Y-%m-%d %H:%M')}\n"
+            f"**Generated**: {datetime.now(timezone.utc).strftime('%Y-%m-%dT%H:%M:%SZ')}\n"
         )
+        if tool_version:
+            header += f"**Tool Version**: `{tool_version}`\n"
+        if profile:
+            header += f"**Profile**: `{profile}`\n"
         if target_url:
             header += f"**Target**: `{target_url}`\n"
         if model:
             header += f"**Model**: `{model}`\n"
+        if tool_commit:
+            header += f"**Tool Commit**: `{tool_commit}`\n"
 
         header += "\n## Risk Summary\n\n"
         for level, msg in self.summary:

audit.py

@@ -4,8 +4,8 @@
 # Regenerate after modular audit changes with:
 #   python3 scripts/build-standalone.py
 # CI verifies this generated artifact plus key behavior regressions.
-# source_sha256: 57cc4ddc3ccb3ed54e76e82dfc447c6e21322c2c03ab804625924e7ec655f245
-# standalone_body_sha256: 3adf747824d6eeb9df837a1e6ae8fd6fe5aa861c57b6a422b9367e52ee0c630f
+# source_sha256: 0948426703dede1934aa265e5558c50f9d993930e2c41b1b444deff9be4b36e2
+# standalone_body_sha256: ec1ca8ab319f24698d522d7e5651145d9e25b4a096921d65a5bad72582de8882
 # END GENERATED STANDALONE HEADER
 
 """
@@ -1624,7 +1624,7 @@ def iter_stdout():
 
 """Markdown report generator for audit results."""
 
-from datetime import datetime
+from datetime import datetime, timezone
 
 
 class Reporter:
@@ -1701,7 +1701,8 @@ def flag(self, level, msg):
         self.summary.append((level, msg))
         self.sections.append(f"{icon} **{msg}**\n")
 
-    def render(self, target_url="", model=""):
+    def render(self, target_url="", model="", tool_version="", profile="",
+               tool_commit=""):
         """Render the complete Markdown report.
 
         Produces a header block (title, metadata, risk summary) followed
@@ -1712,6 +1713,11 @@ def render(self, target_url="", model=""):
                 metadata when provided.
             model: The model identifier used for the audit. Shown in the
                 report metadata when provided.
+            tool_version: API Relay Audit version used for the run.
+            profile: Audit profile used for the run (``general``, ``web3``,
+                or ``full``).
+            tool_commit: Optional git commit for checkout-based runs. Omitted
+                when the standalone script is run outside a repository.
 
         Returns:
             A single Markdown string containing the full report.
@@ -1724,12 +1730,18 @@ def render(self, target_url="", model=""):
         """
         header = (
             f"# API Relay Security Audit Report\n\n"
-            f"**Generated**: {datetime.now().strftime('%Y-%m-%d %H:%M')}\n"
+            f"**Generated**: {datetime.now(timezone.utc).strftime('%Y-%m-%dT%H:%M:%SZ')}\n"
         )
+        if tool_version:
+            header += f"**Tool Version**: `{tool_version}`\n"
+        if profile:
+            header += f"**Profile**: `{profile}`\n"
         if target_url:
             header += f"**Target**: `{target_url}`\n"
         if model:
             header += f"**Model**: `{model}`\n"
+        if tool_commit:
+            header += f"**Tool Commit**: `{tool_commit}`\n"
 
         header += "\n## Risk Summary\n\n"
         for level, msg in self.summary:
@@ -4839,6 +4851,26 @@ def run_channel_classifier(client):
 
 
 
+TOOL_VERSION_FALLBACK = "2.3.0"
+
+
+def _api_relay_audit_checkout_root(script_path):
+    """Return this project's checkout root, or ``None`` for copied scripts."""
+    script_path = script_path.resolve()
+    candidates = []
+    if script_path.parent.name == "scripts":
+        candidates.append(script_path.parent.parent)
+    candidates.append(script_path.parent)
+
+    for root in candidates:
+        if all((
+            (root / "VERSION").is_file(),
+            (root / "scripts" / "build-standalone.py").is_file(),
+            (root / "api_relay_audit" / "reporter.py").is_file(),
+        )):
+            return root
+    return None
+
 
 def _format_identity_inconsistency(non_claude_matches):
     """Render Step 5's non-Claude self-ID finding without over-attribution."""
@@ -4866,6 +4898,51 @@ def _report_error(report, error, status=None):
     report.p(format_diagnosis(_diagnosis_for_error(error, status=status)))
 
 
+def _tool_version():
+    """Return the packaged tool version for report metadata."""
+    repo_root = _api_relay_audit_checkout_root(Path(__file__).resolve())
+    if repo_root is not None:
+        candidate = repo_root / "VERSION"
+        try:
+            value = candidate.read_text(encoding="utf-8").strip()
+        except OSError:
+            value = ""
+        if re.fullmatch(r"\d+\.\d+\.\d+", value):
+            return value
+    return TOOL_VERSION_FALLBACK
+
+
+def _tool_commit_from_checkout():
+    """Return a short git commit only when this script is in this repo checkout."""
+    repo_root = _api_relay_audit_checkout_root(Path(__file__).resolve())
+    if repo_root is None:
+        return ""
+    if not (repo_root / ".git").exists():
+        return ""
+    try:
+        root_result = subprocess.run(
+            ["git", "-C", str(repo_root), "rev-parse", "--show-toplevel"],
+            capture_output=True,
+            text=True,
+            timeout=2,
+            check=True,
+        )
+        git_root = Path(root_result.stdout.strip()).resolve()
+        if git_root != repo_root.resolve():
+            return ""
+        result = subprocess.run(
+            ["git", "-C", str(repo_root), "rev-parse", "--short=12", "HEAD"],
+            capture_output=True,
+            text=True,
+            timeout=2,
+            check=True,
+        )
+    except Exception:
+        return ""
+    commit = result.stdout.strip()
+    return commit if re.fullmatch(r"[0-9a-f]{7,12}", commit) else ""
+
+
 # ============================================================
 # CLI
 # ============================================================
@@ -6688,7 +6765,13 @@ def main():
                  "anomaly, or Web3 injection detected.")
 
     # Output
-    md = report.render(target_url=client.base_url, model=args.model)
+    md = report.render(
+        target_url=client.base_url,
+        model=args.model,
+        tool_version=f"v{_tool_version()}",
+        profile=args.profile,
+        tool_commit=_tool_commit_from_checkout(),
+    )
 
     if args.output:
         Path(args.output).parent.mkdir(parents=True, exist_ok=True)

docs/_metrics.md

@@ -16,23 +16,23 @@
 | 单文件版版本 | `v2.3` | `audit.py` docstring |
 | 步骤数 (Step N) | **14** | grep `Step N` in `scripts/audit.py` |
 | 步骤数 (单文件版) | 14 | grep `Step N` in `audit.py` |
-| 测试数 (pytest) | **778** | `pytest --collect-only` |
-| 测试数 (static) | 754 | grep `def test_*` in tests/ |
+| 测试数 (pytest) | **782** | `pytest --collect-only` |
+| 测试数 (static) | 758 | grep `def test_*` in tests/ |
 | CLI flag 数 | 21 | grep `add_argument("--*")` |
 | profile 选项 | general, web3, full | argparse choices |
 | ROADMAP 上次更新 | 2026-06-07 | `ROADMAP.md` 头部 |
 | Codex review 提及次数 | 4 | grep `Codex review (cycle\|round)` 在 Shipped 节 |
 | Codex review 已编号轮次（最大） | 6 | grep `Nth Codex review round` |
 | Codex bug 累计（最新声称） | 18 | grep `cumulative N real bug` |
-| 测试数演进 (ROADMAP) | [546, 560, 562, 586, 642, 700, 778] | grep `Final test count: N/N passing` |
-| Recorded commit SHA | `eeb9a2c` | recent reachable commit; `--check` allows follow-up metrics commits |
-| Recorded commit date | 2026-06-07 | recent reachable commit; `--check` allows follow-up metrics commits |
+| 测试数演进 (ROADMAP) | [546, 560, 562, 586, 642, 700, 782] | grep `Final test count: N/N passing` |
+| Recorded commit SHA | `68d4141` | recent reachable commit; `--check` allows follow-up metrics commits |
+| Recorded commit date | 2026-06-12 | recent reachable commit; `--check` allows follow-up metrics commits |
 
 ## 一致性自检
 
 - ✅ 版本一致：两份都是 `v2.3`。
 - ✅ 步骤数一致：14。
-- ℹ️ pytest (778) vs 静态 (754) 差距 >20，多出来的来自 parametrize/fixture——以 pytest 为准。
+- ℹ️ pytest (782) vs 静态 (758) 差距 >20，多出来的来自 parametrize/fixture——以 pytest 为准。
 
 ## 人工 review 边界（脚本抓不到，每次发布要人工核对）
 

scripts/audit.py

@@ -69,6 +69,26 @@
 from api_relay_audit.tool_substitution import run_tool_substitution_test
 from api_relay_audit.web3.injection_probes import run_web3_injection_probes
 
+TOOL_VERSION_FALLBACK = "2.3.0"
+
+
+def _api_relay_audit_checkout_root(script_path):
+    """Return this project's checkout root, or ``None`` for copied scripts."""
+    script_path = script_path.resolve()
+    candidates = []
+    if script_path.parent.name == "scripts":
+        candidates.append(script_path.parent.parent)
+    candidates.append(script_path.parent)
+
+    for root in candidates:
+        if all((
+            (root / "VERSION").is_file(),
+            (root / "scripts" / "build-standalone.py").is_file(),
+            (root / "api_relay_audit" / "reporter.py").is_file(),
+        )):
+            return root
+    return None
+
 
 def _format_identity_inconsistency(non_claude_matches):
     """Render Step 5's non-Claude self-ID finding without over-attribution."""
@@ -96,6 +116,51 @@ def _report_error(report, error, status=None):
     report.p(format_diagnosis(_diagnosis_for_error(error, status=status)))
 
 
+def _tool_version():
+    """Return the packaged tool version for report metadata."""
+    repo_root = _api_relay_audit_checkout_root(Path(__file__).resolve())
+    if repo_root is not None:
+        candidate = repo_root / "VERSION"
+        try:
+            value = candidate.read_text(encoding="utf-8").strip()
+        except OSError:
+            value = ""
+        if re.fullmatch(r"\d+\.\d+\.\d+", value):
+            return value
+    return TOOL_VERSION_FALLBACK
+
+
+def _tool_commit_from_checkout():
+    """Return a short git commit only when this script is in this repo checkout."""
+    repo_root = _api_relay_audit_checkout_root(Path(__file__).resolve())
+    if repo_root is None:
+        return ""
+    if not (repo_root / ".git").exists():
+        return ""
+    try:
+        root_result = subprocess.run(
+            ["git", "-C", str(repo_root), "rev-parse", "--show-toplevel"],
+            capture_output=True,
+            text=True,
+            timeout=2,
+            check=True,
+        )
+        git_root = Path(root_result.stdout.strip()).resolve()
+        if git_root != repo_root.resolve():
+            return ""
+        result = subprocess.run(
+            ["git", "-C", str(repo_root), "rev-parse", "--short=12", "HEAD"],
+            capture_output=True,
+            text=True,
+            timeout=2,
+            check=True,
+        )
+    except Exception:
+        return ""
+    commit = result.stdout.strip()
+    return commit if re.fullmatch(r"[0-9a-f]{7,12}", commit) else ""
+
+
 # ============================================================
 # CLI
 # ============================================================
@@ -1919,7 +1984,13 @@ def main():
                  "anomaly, or Web3 injection detected.")
 
     # Output
-    md = report.render(target_url=client.base_url, model=args.model)
+    md = report.render(
+        target_url=client.base_url,
+        model=args.model,
+        tool_version=f"v{_tool_version()}",
+        profile=args.profile,
+        tool_commit=_tool_commit_from_checkout(),
+    )
 
     if args.output:
         Path(args.output).parent.mkdir(parents=True, exist_ok=True)

scripts/sync-version.py

@@ -103,12 +103,18 @@ def replace_regex(
 
 def sync_audit_script(version: Version, path: Path) -> str:
     text = read_text(path)
-    return replace_regex(
+    text = replace_regex(
         text,
         rf"API Relay Security Audit Tool {DISPLAY_VERSION_RE}",
         f"API Relay Security Audit Tool {version.display}",
         path,
     )
+    return replace_regex(
+        text,
+        rf'^TOOL_VERSION_FALLBACK = "{FULL_VERSION_RE}"$',
+        f'TOOL_VERSION_FALLBACK = "{version.full}"',
+        path,
+    )
 
 
 def sync_build_standalone(version: Version, path: Path) -> str: