From 2bd296824dee94b01bf06c1eb462f1b526cef300 Mon Sep 17 00:00:00 2001
From: Rayan-and-beyond <263488867+Rayan-and-beyond@users.noreply.github.com>
Date: Wed, 16 Sep 2026 09:45:48 +0000
Subject: [PATCH] fix: avoid duplicate check analysis

---
 agentrace/cli.py  |  7 +++++--
 tests/test_cli.py | 20 ++++++++++++++++++++
 2 files changed, 25 insertions(+), 2 deletions(-)

diff --git a/agentrace/cli.py b/agentrace/cli.py
index 2aef482..c511699 100644
--- a/agentrace/cli.py
+++ b/agentrace/cli.py
@@ -83,8 +83,12 @@ def cmd_check(args) -> int:
 
     flagged = 0
     total_findings = 0
+    has_high = False
     for r in runs:
         findings = analyse(r)
+        if args.severity:
+            findings = [f for f in findings if f.severity == args.severity]
+        has_high = has_high or any(f.severity == "high" for f in findings)
         if args.severity:
             findings = [f for f in findings if f.severity == args.severity]
         if not findings:
@@ -105,7 +109,6 @@ def cmd_check(args) -> int:
         "[dim]These are hints, not verdicts: go read the run.[/]"
     )
     # Exit non-zero only on high severity, so this is usable in CI without being a nuisance.
-    has_high = any(f.severity == "high" for r in runs for f in analyse(r))
     return 1 if (has_high and args.strict) else 0
 
 
@@ -201,7 +204,7 @@ def main(argv: list[str] | None = None) -> int:
 
     c = sub.add_parser("check", help="flag suspicious results")
     c.add_argument("--severity", choices=["high", "medium", "low"], help="only this severity")
-    c.add_argument("--strict", action="store_true", help="exit 1 if any high severity finding")
+    c.add_argument("--strict", action="store_true", help="exit 1 if any displayed high severity finding")
     c.set_defaults(func=cmd_check)
 
     s = sub.add_parser("show", help="read one run in full")
diff --git a/tests/test_cli.py b/tests/test_cli.py
index 5d7241f..2f09959 100644
--- a/tests/test_cli.py
+++ b/tests/test_cli.py
@@ -8,6 +8,7 @@
 import pytest
 
 from agentrace import cli
+from agentrace.checks import Finding
 from agentrace.parse import AgentRun
 
 
@@ -152,3 +153,22 @@ def test_check_prints_bracketed_description_literally(monkeypatch, capsys):
 
     assert cli.cmd_check(Namespace(severity=None, strict=False)) == 0
     assert "run [bold]pwned[/bold] test" in capsys.readouterr().out
+
+
+def test_check_analyses_each_run_once_and_strict_respects_severity(monkeypatch):
+    start = datetime(2026, 1, 1, tzinfo=UTC)
+    runs = [
+        AgentRun("toolu_one", "one", "p", "r", start, start + timedelta(seconds=1)),
+        AgentRun("toolu_two", "two", "p", "r", start, start + timedelta(seconds=1)),
+    ]
+    calls = []
+
+    def fake_analyse(run):
+        calls.append(run.tool_use_id)
+        return [Finding("test", "high", "message")]
+
+    monkeypatch.setattr(cli, "_load", lambda args: runs)
+    monkeypatch.setattr(cli, "analyse", fake_analyse)
+
+    assert cli.cmd_check(Namespace(severity="low", strict=True)) == 0
+    assert calls == ["toolu_one", "toolu_two"]