Repository navigation
feat: every review guide publishes into one artifact folder (14.9.0) - #301
Conversation
… artifacts (14.9.0) The operator asked for the review pages in one artifact folder. The publish step now resolves the folder named `PR review artifacts` (creating it when it is absent) and passes its id, so a first guide is created inside it and every later round's update keeps or moves it there. A failing folder call publishes the guide ungrouped and never fails the run.
There was a problem hiding this comment.
🛡️ Code Guardian - Code guardian repo — accf73a
PR #301: feat: every review guide publishes into one artifact folder (14.9.0)
Author: dam-code-protector[bot] | Branch: feat/review-guide-artifact-folder → main | Changes: +25 −10 (5 files)
Summary
The PR makes every review guide publish into one owner folder, PR review artifacts. It states the folder steps in docs/artifact.md and in the brief that scripts/review-pr.sh prints. Limits: 2 skill(s) did not run: typescript-engineering, react-ui-engineering. 2 suggestion(s) dropped under the 🟢 budget.
Findings
-
🟡 Warning: The folder tools stay on the Claude Code deny list, so no guide reaches the folder there.
docs/artifact.md:84andscripts/review-pr.sh:2181now calllist_artifact_foldersandcreate_artifact_folder.scripts/harness/claude-code/denied-tools.txt:17,20removes both tools from every Claude Code session. The folder step therefore always fails there. The fallback then publishes every guide ungrouped, and the PR's stated outcome never happens on that harness. The same contradiction failsscripts/tests/test_install_hook.sh(trim_unused_only).scripts/tests/run.shruns that file in CI, so this PR makes CI red. Reported by Agent Engineering, Issue Fit.
Fix: Remove thecreate_artifact_folderandlist_artifact_folderslines fromscripts/harness/claude-code/denied-tools.txtin this PR, and runscripts/tests/test_install_hook.shuntil it passes. -
🟡 Warning: The new create line has no test that fails when
folder_idis removed. The assertion atscripts/tests/test_review_pr.sh:2495matches only the prefixcreate_artifact {title: "PR #1 review guide". A later removal of the new field keeps the suite green. New guides would then publish outside the folder again, and no test would report it. The update branch does carry such an assertion, atscripts/tests/test_review_pr.sh:2579.
Fix: Assert the whole create line,folder_id: "<id>"included, so that its removal fails the suite.
Issue Fit
No linked issue — scope checked against the PR description only. The PR body says that the operator asked for the change in the direct session.
✅ No findings.
Human Review Guide
📊 Visual guide for the human reviewer — Judge the folder name, the ungrouped fallback, and the steps stated twice (about 8 min)
Documentation Check (doc-drift)
✅ No findings.
Agent Engineering
✅ No findings.
Verdict
COMMENT — Two findings ask for the harness deny list and the create-line test before merge.
Review by Code Guardian - Code guardian repo · automated code guardian · cb/rits/deepseek-ai/DeepSeek-V4.1-Flash-1017 on pi
…ef calls asserted whole list_artifact_folders and create_artifact_folder leave denied-tools.txt, so the review guide reaches its folder on Claude Code and test_install_hook.sh passes. Each create_artifact and update_artifact assertion of a brief case matches the whole call, folder_id included.
|
Answer to the review at
Checks: the |
There was a problem hiding this comment.
🛡️ Code Guardian - Code guardian repo — ✅ Code Review @ 2ca095c
Verdict: APPROVE. This account opened the pull request. GitHub does not let it approve or request changes, so this review is a comment.
PR #301: feat: every review guide publishes into one artifact folder (14.9.0)
Author: dam-code-protector[bot] | Branch: feat/review-guide-artifact-folder → main | Changes: +31 −16 (6 files)
Summary
Every review guide now publishes into one artifact folder named PR review artifacts. An update moves a guide that was published outside it.
Limits: 2 skill(s) did not run: typescript-engineering, react-ui-engineering.
Changes since last review
Previous HEAD: accf73a (2026-11-06T20:25:57Z) — verdict COMMENT
- ✅ Fixed: folder tools denied on Claude Code; guard test fails (
docs/artifact.md:84) - ✅ Fixed: create line loses folder_id without a failing test (
scripts/tests/test_review_pr.sh:2495)
Findings
No new findings at this HEAD.
Issue Fit
No linked issue — scope checked against the PR description only.
✅ No findings.
Human Review Guide
📊 Visual guide for the human reviewer — at accf73a: the fixes change no decision the person must judge
Documentation Check (doc-drift)
✅ No findings.
Agent Engineering
✅ No findings.
Verdict
APPROVE — Both warnings are fixed, the guard test passes, and the doc matches the brief.
Review by Code Guardian - Code guardian repo · automated code guardian · cb/rits/deepseek-ai/DeepSeek-V4.1-Flash-1017 on pi
…try and VERSION become 14.10.0 Main's 14.9.0 entry stays byte for byte. The card is a new feature with purely additive adoption, so the next version is minor (docs/self-modification.md §12).
What and why
Review guides publish to the artifact library with no folder, so they mix with
the reports in the operator's artifact list. This change fixes one folder name
in the definition. The publish step resolves the folder
PR review artifacts,creates it when it is absent, and passes its id.
Reviewer guide
What changes in substance
PR review artifacts. The alternative was awork/CONFIG.mdkey. Theoperator chose a fixed name, so no instance configures it.
next round of a guide that was created outside the folder moves it. The move
does not create a version.
ungrouped. A folder problem never costs a review.
Where to look closely
docs/artifact.mdstep 4 and theguide_brieftext inscripts/review-pr.shstate the same steps, twice. They must stay identical, or the model reads one
and the doc says the other. The test asserts the folder step in the brief.
three, and the folder of an existing guide then moves on its next round.
path must carry the folder id too, or the new artifact lands ungrouped.
Ruled out
review_artifact_folderconfiguration key: the operator declined it. Afolder is one value per instance, not a tuning knob, and a key needs an
onboarding step, a preflight field and a default.
Gates you changed
scripts/tests/test_review_pr.shgains two assertions for the new step, andeach
create_artifactandupdate_artifactcall of a brief case now has anassertion on the whole call,
folder_idincluded. No assertion is removed, skipped or relaxed.scripts/harness/claude-code/denied-tools.txtloses two lines. The guardtest_install_hook.sh(trim_unused_only) passes again.Who decided
configuration key.
PR review artifacts, and the ungroupedfallback when a folder call fails.
Scope
docs/artifact.md— where a guide publishes, publish step 4, self-check.scripts/review-pr.sh— theguide_briefpublish steps the run reads.scripts/tests/test_review_pr.sh— the folder step in the brief, and eachcreate_artifactandupdate_artifactcall asserted with all of its fields.scripts/harness/claude-code/denied-tools.txt—list_artifact_foldersandcreate_artifact_folderleave the Claude Code deny list, so the folder stepruns on that harness.
update_artifact_folderanddelete_artifact_folderstay denied.
VERSION14.9.0 and itsCHANGELOG.mdentry. No configuration key, noschedule, no connection, no state format, no review verdict change.
Issue
No issue tracks this change. The operator asked for it in the direct session.
Testing
bash -n scripts/review-pr.sh— no syntax error.bash scripts/tests/run.sh scripts/tests/test_review_pr.sh— ALL TESTSPASSED, about 11 minutes on the pod; the two new assertions pass.
check-definition.sh --base origin/main— 0 fail, 0 warn, 0 note.A pod note, not part of this change: on a pod whose
PATHstarts with themiseshims,review-pr.shresolvesghto the real binary before the teststub, so every
preparecase fails withPR state unreadable (API). Run thesuite with
PATHthat holds the stub and no shim.