Fix skip heuristics dropping real findings, delta budget double-count

P1: Remove _should_skip_line() from block-start matching. If a line matches
_BLOCK_START, it's a finding — trust the severity pattern regardless of
summary content. Skip heuristics remain in the uncertainty scanner only.
Prevents findings containing "Path to Approval", "Looks good", etc. in
their summaries from being silently dropped.

P2: Fix delta-mode token budget to not count changed_files_text (which is
NOT included in delta-mode prompt, only in fresh reviews). Prevents
unnecessary import-context dropping in re-reviews.

P2: Add 2 regression tests for findings with skip markers in summaries.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: igerber

.claude/scripts/openai_review.py

@@ -563,9 +563,10 @@ def parse_review_findings(
                 current_section = heading
             continue
 
-        # Check for block start
+        # Check for block start — trust the severity pattern; don't apply
+        # skip heuristics here (they're for the uncertainty scanner only)
         sev_match = _BLOCK_START.search(line)
-        if sev_match and not _should_skip_line(line):
+        if sev_match:
             # Flush previous block
             if current_block is not None:
                 blocks.append((current_severity, current_block_section, current_block))
@@ -1473,8 +1474,11 @@ def main() -> None:
         estimate_tokens(criteria_text)
         + estimate_tokens(registry_content)
         + estimate_tokens(diff_text)
-        + estimate_tokens(changed_files_text)
     )
+    # In delta mode, changed_files_text is NOT in the prompt (replaced by delta sections);
+    # only count it for fresh reviews where it appears in "Changes Under Review"
+    if not delta_diff_text:
+        mandatory_est += estimate_tokens(changed_files_text)
     if previous_review:
         mandatory_est += estimate_tokens(previous_review)
     if delta_diff_text:

tests/test_openai_review.py

@@ -951,6 +951,21 @@ def test_parses_plain_label_format(self, review_mod):
         assert len(findings) == 1
         assert findings[0]["severity"] == "P2"
 
+    def test_finding_with_skip_marker_in_summary_still_parsed(self, review_mod):
+        """Findings whose summaries contain skip markers like 'Path to Approval' should parse."""
+        review_text = "**P2** The prompt omits the Path to Approval section in `foo.py:L10`\n"
+        findings, uncertain = review_mod.parse_review_findings(review_text, 1)
+        assert len(findings) == 1
+        assert findings[0]["severity"] == "P2"
+        assert not uncertain
+
+    def test_finding_with_looks_good_in_summary(self, review_mod):
+        """Finding mentioning 'Looks good' in summary should not be skipped."""
+        review_text = "**P1** Assessment says Looks good but edge case is unhandled in `bar.py:L5`\n"
+        findings, _ = review_mod.parse_review_findings(review_text, 1)
+        assert len(findings) == 1
+        assert findings[0]["severity"] == "P1"
+
     def test_parses_multiline_finding_block(self, review_mod):
         """Multi-line finding blocks (Severity/Impact on separate lines)."""
         review_text = (