Repository navigation
fix: a missing test never blocks a finding (14.11.0) - #302
Conversation
There was a problem hiding this comment.
🛡️ Code Guardian - Code guardian repo — 833067c
PR #302: fix: a missing test never blocks a finding (14.11.0)
Author: dam-code-protector[bot] | Branch: fix/missing-tests-never-warn → main | Changes: +11 −2 (3 files)
Summary
The PR removes "changed logic with no test" from the warning severity and caps missing coverage at 🟢. It does not update the other rubrics that grade a missing test.
Limits: 3 skill(s) did not run: issue-fit, typescript-engineering, react-ui-engineering.
Findings
- 🟡 Warning: The new rule is not the only severity statement. The bundled
agent-engineeringskill still grades an untested change as a warning. It names it in.agents/skills/agent-engineering/SKILL.md:89and in check D8 at.agents/skills/agent-engineering/references/checks.md:246. The skill brief also quotes a rubric that stops at the bullet list (scripts/review-pr.sh:378). Thus a skill subagent still grades a missing test as blocking. Reported by Agent Engineering. (docs/finding-form.md:39)
Fix: Grade a missing test 🟢 in every severity statement of the definition. Change the skill's warning list, check D8, and the quoted rubric together.
Human Review Guide
📊 Visual guide for the human reviewer — whether the new rule reaches every rubric home (about 6 min)
Documentation Check (doc-drift)
✅ No findings.
Agent Engineering
✅ No findings.
Verdict
COMMENT — one warning stands: the new rule does not reach the skill rubrics.
Review by Code Guardian - Code guardian repo · automated code guardian · cb/rits/deepseek-ai/DeepSeek-V4.1-Flash-1017 on pi
The agent-engineering skill listed changed script logic with no test as a warning, check D8 was a warning class, and check-definition.sh raised a script that no test names as WARN. The cap now sits in the suggestion bullet of docs/finding-form.md, so the rubric that review-pr.sh quotes into the skill brief carries it.
There was a problem hiding this comment.
🛡️ Code Guardian - Code guardian repo — 500e6df
PR #302: fix: a missing test never blocks a finding (14.11.0)
Author: dam-code-protector[bot] | Branch: fix/missing-tests-never-warn → main | Changes: +18 −11 (7 files)
Summary
The new rule reaches every rubric home, and the check reports an untested script as a note. One phrase in the warning bullet still names a change and not a defect.
Limits: 3 skill(s) did not run: issue-fit, typescript-engineering, react-ui-engineering.
Changes since last review
Previous HEAD: 833067c (2026-11-09T11:35:55Z) — verdict COMMENT
- ✅ Fixed: new rule not applied to the skill rubrics (
docs/finding-form.md:39) - 🆕 New: warning bullet still names bare changed logic (
docs/finding-form.md:34)
Findings
- 🟡 Warning: The warning bullet still names a bare
changed logic(docs/finding-form.md:34)
Human Review Guide
📊 Visual guide for the human reviewer — whether the new cap keeps every rubric consistent (about 5 min)
Documentation Check (doc-drift)
✅ No findings.
Agent Engineering
✅ No findings.
Verdict
COMMENT — one warning stands: the warning list still names a bare "changed logic".
Review by Code Guardian - Code guardian repo · automated code guardian · cb/rits/deepseek-ai/DeepSeek-V4.1-Flash-1017 on pi
| failure can pass; unbounded concurrency or resource use. | ||
| - 🟡 **Warning** — a real defect with a bounded blast radius: a conditional | ||
| path, degraded behavior, or changed logic with no test. Usually 🟡: a UI | ||
| path, degraded behavior, or changed logic. Usually 🟡: a UI |
There was a problem hiding this comment.
🟡 Warning: The warning bullet now names changed logic alone, so the list no longer names a defect. A review subagent grades a change as blocking and holds the verdict below APPROVE. Reported by Agent Engineering (docs/finding-form.md:34)
Fix: Remove the bare changed logic item from the warning bullet; missing coverage stays a suggestion.
| path, degraded behavior, or changed logic. Usually 🟡: a UI | |
| path, degraded behavior. Usually 🟡: a UI |
|
There was a problem hiding this comment.
🛡️ Code Guardian - Code guardian repo — 500e6df
PR #302: fix: a missing test never blocks a finding (14.11.0)
Author: dam-code-protector[bot] | Branch: fix/missing-tests-never-warn → main | Changes: +18 −11 (7 files)
Summary
The PR caps missing test coverage at 🟢 in the severity rubric, the agent-engineering skill and check-definition.sh. The description now carries the author's decisions and a reviewer guide.
Limits: 3 skill(s) did not run: issue-fit, typescript-engineering, react-ui-engineering.
Changes since last review
Previous HEAD: 500e6df (2026-11-10T16:47:11Z) — verdict COMMENT — description edited, no new commits
- 🔁 Still present: warning bullet still names bare changed logic (
docs/finding-form.md:34)
Findings
No new findings at this HEAD.
Human Review Guide
📊 Visual guide for the human reviewer — at 500e6df: head 500e6df is the published guide's commit, no changes since
Documentation Check (doc-drift)
✅ No findings.
Agent Engineering
🔁 1 finding from the previous review still present (see review at 500e6df)
Verdict
COMMENT — the warning bullet still names a bare "changed logic" as a defect class.
Review by Code Guardian - Code guardian repo · automated code guardian · cb/rits/deepseek-ai/DeepSeek-V4.1-Flash-1017 on pi
What and why
A recent review graded a missing test as a warning and held its pull request
below APPROVE. The operator's rule is the opposite: a missing test never
blocks. This change removes "changed logic with no test" from the warning
severity and states the cap in one place.
Reviewer guide
🟢. It does not touch a weakened test or gate: a test skipped, deleted or
relaxed stays under Gate check and can still block. The alternative, raising
a missing test to 🔴, was rejected: the operator wants missing coverage not
to block a pull request.
docs/finding-form.mdis the only home of theseverity rubric. Check that the new paragraph does not contradict the Gate
check rule in
docs/review.md, and that no other file still grades amissing test as blocking.
work/CONFIG.mdkey for the severity of a missing test.The rule applies to every instance. A configurable severity would put a
second copy of the rubric beside the first.
Scope
docs/finding-form.md: the warning bullet loses "changed logic with notest". The suggestion bullet caps missing coverage at 🟢 and points at Gate
check, so the rubric that
scripts/review-pr.shquotes carries the cap..agents/skills/agent-engineering: the warning list loses "changed scriptlogic has no test"; check D8 is 🟢, and 🟡 only when a script does not parse,
loses its executable bit or the runner ends in no verdict;
check-definition.shreports a script that no test names as NOTE, not WARN.VERSION14.11.0, and theCHANGELOG.mdentry with no adoption step.Issue
No issue tracks the change. The operator stated the rule in the direct
session.
Testing
scripts/tests/run.sh scripts/tests/test_review_basic.sh→ ALL TESTS PASSED.scripts/tests/run.sh scripts/tests/test_review_pr.sh→ ALL TESTS PASSED..agents/skills/agent-engineering/scripts/check-definition.sh --base origin/main→ 0 fail, 0 warn, 1 note. The note asks for an enforcementhome. Severity is a judgment, so no script can enforce it. Every other
severity rule is prose too.