From 83335639dd5798a7a9b739f2b3b9a1422049ff8d Mon Sep 17 00:00:00 2001
From: datacas <4690232+datacas@users.noreply.github.com>
Date: Fri, 30 Oct 2026 23:35:06 +0100
Subject: [PATCH] fix(verification): reject malformed check entries

---
 docs/verification.md            |  5 +++++
 scripts/verification_plan.py    |  8 ++++++++
 skills/cc-verify/SKILL.md       |  3 +++
 tests/test_verification_plan.py | 27 +++++++++++++++++++++++++++
 4 files changed, 43 insertions(+)

diff --git a/docs/verification.md b/docs/verification.md
index e253050..373028b 100644
--- a/docs/verification.md
+++ b/docs/verification.md
@@ -236,6 +236,11 @@ top-level `tests.evidence` list. This is separate from the unchanged code-host
 }
 ```
 
+Every entry in a check's `evidence` and `attempts` arrays must be a JSON
+object. A non-object entry (including a string) is malformed; the runtime
+rejects the report and caps the conclusion at `not_verified` rather than
+coercing text into evidence.
+
 Freeze each check before execution. A result names the same check ID, command,
 executor/profile and context. The runtime supplies frozen check identities to
 later code stages. Copy that identity, including its owning executor/profile,
diff --git a/scripts/verification_plan.py b/scripts/verification_plan.py
index cf28e7e..8179b01 100644
--- a/scripts/verification_plan.py
+++ b/scripts/verification_plan.py
@@ -342,6 +342,14 @@ def normalize_tests(tests: Mapping[str, Any], head_sha: str, *,
             prior = CheckResult.from_dict({**previous, "id": check.id})
             if prior not in plan.results:
                 plan = plan.record(prior)
+        for field in ("evidence", "attempts"):
+            values = entry.get(field, ())
+            if not isinstance(values, (list, tuple)):
+                raise VerificationError(f"check {check.id} {field} must be an array")
+            for index, item in enumerate(values):
+                if not isinstance(item, Mapping):
+                    raise VerificationError(
+                        f"check {check.id} {field}[{index}] must be an object, got {type(item).__name__}")
         result = CheckResult(check.id, entry["head_sha"], entry["result"],
                              tuple(entry.get("evidence", ())), entry.get("cause", ""),
                              tuple(entry.get("attempts", ())))
diff --git a/skills/cc-verify/SKILL.md b/skills/cc-verify/SKILL.md
index db6b661..47f2599 100644
--- a/skills/cc-verify/SKILL.md
+++ b/skills/cc-verify/SKILL.md
@@ -28,6 +28,9 @@ obligations.
 Each result also carries `head_sha`, `result: pass|fail|unverified`,
 `evidence` (integer `exit_code`, `executed` count or null), `cause` and
 `attempts`. The enclosing stage result names its current `head_sha` too.
+Each `evidence` and `attempts` array entry must be a JSON object. Do not send
+text entries or rely on the runtime to convert them; malformed entries make
+the report invalid and cap its conclusion at `not_verified`.
 An unexecuted check is explicitly `unverified` with a cause and the recovery
 attempts actually made; do not invent an exit code. Keep the check's identity
 when rerunning it and retain previous results. Never replace a local
diff --git a/tests/test_verification_plan.py b/tests/test_verification_plan.py
index 4f2dde1..06364e2 100644
--- a/tests/test_verification_plan.py
+++ b/tests/test_verification_plan.py
@@ -220,6 +220,33 @@ def test_invalid_followup_never_raises_failed_agent_or_frozen_conclusion(self):
                 "version": 99, "conclusion": conclusion, "passed": True}}, plan)
             self.assertEqual(rc._legacy_tests(payload)["verification"], "failed")
 
+    def test_v2_rejects_string_evidence_and_attempts_with_check_name(self):
+        plan = vp.VerificationPlan(criteria()).extend((check(),))
+        for field in ("evidence", "attempts"):
+            entry = report(**{field: ["text"]})
+            payload = {"version": 2, "conclusion": "verified", "checks": [entry]}
+            with self.subTest(field=field), self.assertRaisesRegex(
+                    vp.VerificationError, rf"check check-1 {field}\[0\].*str"):
+                vp.normalize_tests(payload, HEAD, plan=plan)
+
+    def test_v1_string_evidence_rejects_with_plan_and_stays_unchanged_without_one(self):
+        legacy = {"version": 1, "conclusion": "verified", "evidence": ["text"]}
+        plan = vp.VerificationPlan(criteria()).extend((check(),))
+        with self.assertRaisesRegex(vp.VerificationError, "invalid legacy evidence"):
+            vp.normalize_tests(legacy, HEAD, plan=plan)
+        unchanged, updated = vp.normalize_tests(legacy, HEAD)
+        self.assertEqual(unchanged, legacy)
+        self.assertIsNone(updated)
+
+    def test_string_check_entries_reach_cycle_as_not_verified(self):
+        for field in ("evidence", "attempts"):
+            payload, _ = rc._normalize_test_payload({"head_sha": HEAD, "tests": {
+                "version": 2, "conclusion": "verified", "passed": True,
+                "checks": [report(**{field: ["text"]})]}})
+            with self.subTest(field=field):
+                self.assertEqual(payload["tests"]["conclusion"], "not_verified")
+                self.assertFalse(payload["tests"]["passed"])
+
     def test_legacy_unmatched_failures_are_retained(self):
         plan = vp.VerificationPlan(criteria()).extend((check(),))
         payload, updated = rc._normalize_test_payload({"status": "IMPLEMENTED", "head_sha": HEAD,