claude-dev-env 8.92.0 → 8.92.1

package/.agents/skills/e-code-review/SKILL.md

@@ -8,7 +8,7 @@ description: >-
 
 # e-code-review
 
-**Pick a level, load its file, run it.** Each file has the same steps, angles, caps, and fields as the built-in `/code-review` recipe for that level. Only the wording differs, where this repo's prose rules require it.
+**Pick a level, load its file, run it.** Each file has the same steps, angles, caps, and fields as the built-in `/code-review` recipe for that level. Only the wording differs, where this repo's prose rules require it. A question about `/code-review` asks about the built-in recipe, so answer it from the built-in text and use these files only for `/e-code-review`.
 
 | Level | File | Shape |
 |---|---|---|

package/.agents/skills/efficient-review/SKILL.md

@@ -21,7 +21,7 @@ When the plan runtime is `other`, the model is deferred. The running agent appli
 
 ## Loop
 
-Check each finding against the code. Fix every one that holds. Run the review again until a pass reports none. The steps are in [reference/criteria.md](reference/criteria.md).
+Check each finding against the code. Fix every one that holds. Run the review again until a pass reports none. Before a merge, [the pr-lifecycle review loop](../pr-lifecycle/reference/review-loop-before-merge.md) sets how many full passes run. The steps are in [reference/criteria.md](reference/criteria.md).
 
 ## Reading guide
 

package/.agents/skills/efficient-review/reference/criteria.md

@@ -18,4 +18,4 @@ An empty list is a valid review. End with a json block. Each finding has `file`,
 
 ## Loop
 
-Fix every finding that holds, including a file the diff did not touch. Then run the review again. Stop when a pass reports no finding that holds.
+Fix every finding that holds, including a file the diff did not touch. Then run the review again. Stop when a pass reports no finding that holds. Before a merge, [the pr-lifecycle review loop](../../pr-lifecycle/reference/review-loop-before-merge.md) sets how many full passes run.

package/.agents/skills/pr-lifecycle/reference/review-loop-before-merge.md

@@ -2,10 +2,11 @@
 
 A pull request the agent made merges only after a code review of its head comes back clean. Loop it:
 
-1. Review the head with the command `CLAUDE_MERGE_REVIEW_COMMAND` names. Where it is unset, run `/code-review high --max-findings all` against the base.
+1. Review the head with the command `CLAUDE_MERGE_REVIEW_COMMAND` names. Where it is unset, run `/code-review high --max-findings all` against the base. This is a full pass.
 2. Record the pass: `python3 ~/.claude/scripts/merge_review_receipt.py <owner>/<name> <number> <head sha> <finding count>`.
 3. Fix every valid finding, push, and review the new head. Answer each finding you rule out on the pull request with its file:line evidence.
-4. Repeat until a pass records `CLEAN`. A pass whose every finding is ruled out records `CLEAN` with `--ruled-out <finding count> --rulings <link to the rulings comment on this pull request>`.
-5. Merge pinned to that head: `expectedHeadSha` on the merge tool, or `--match-head-commit` on `gh pr merge`.
+4. The first three passes are full passes. Each later pass is `/code-review low --fix <number>`. It reads only the pull request's diff against its base and leaves its fixes uncommitted. Name the number: after a push, a bare `/code-review low` diffs the branch against its own upstream and sees nothing. Commit and push those fixes, then run the next pass on the new head. A pass that reported findings records its finding count, even when `--fix` fixed them all.
+5. Repeat until a pass records `CLEAN`. A pass whose every finding is ruled out records `CLEAN` with `--ruled-out <finding count> --rulings <link to the rulings comment on this pull request>`.
+6. Merge pinned to that head: `expectedHeadSha` on the merge tool, or `--match-head-commit` on `gh pr merge`.
 
 `hooks/blocking/merge_review_gate.py` denies any other merge. In a merge-queue repository, enqueue with `gh pr merge <number> --repo <owner>/<name> --auto --match-head-commit <head sha>`; auto-merge that names no head is denied. A repository whose own CI and merge queue review and merge its pull requests is listed in `CLAUDE_MERGE_REVIEW_EXEMPT_REPOSITORIES`, comma separated as `owner/name`, and skips the gate.

package/hooks/advisory/test_owned_pr_review_loop.py

@@ -281,6 +281,8 @@ def test_should_name_every_step_and_skill_in_the_context() -> None:
     )
     assert "jl-cmd-projects/claude-dev-env#42" in context
     assert "efficient-review" in context
+    assert "three full passes" in context
+    assert "/code-review low --fix 42" in context
     assert "gh pr ready 42 --repo jl-cmd-projects/claude-dev-env" in context
     assert "poteto-mode/playbooks/babysit.md" in context
     assert "poteto-mode/playbooks/shipping.md" in context

package/hooks/blocking/test_merge_review_gate.py

@@ -178,6 +178,12 @@ def test_should_name_the_configured_review_command(monkeypatch: pytest.MonkeyPat
     assert "`/team-review`" in _reason(_merge_tool(CLEAN_HEAD))
 
 
+def test_should_name_the_low_fix_pass_after_three_full_reviews_in_the_deny_reason() -> None:
+    reason = _reason(_merge_tool(CLEAN_HEAD))
+    assert "three full passes" in reason
+    assert "`/code-review low --fix 12`" in reason
+
+
 def test_should_allow_auto_merge_pinned_to_the_clean_head() -> None:
     merge_review_receipt.record_review("acme/tools", 12, CLEAN_HEAD, 0)
     command = f"gh pr merge 12 --repo acme/tools --auto --match-head-commit {CLEAN_HEAD}"

package/hooks/features/conduct-gates.md

@@ -17,7 +17,7 @@ This family stops a chat reply, edited message, or tool call when its payload br
 - `blocking/verify_before_acting.py` blocks a completed mutating call when the agent's reasoning contains an unchecked claim.
 - `blocking/pr_lifecycle_skill_gate.py` denies a commit, push, pull request action, or merge until the `pr-lifecycle` skill appears in the transcript since the last compaction.
 - `blocking/build_eval_before_eval_write_gate.py` denies a write to an eval file until a Skill call named exactly `claude-api` with args `build-eval`, or the `/claude-api build-eval` command, appears in the transcript.
-- `blocking/merge_review_gate.py` denies an agent's merge until a review pass on the merged head reported zero findings, or ruled out every finding and linked the posted rulings. It reads the receipt `scripts/merge_review_receipt.py` writes, requires the merge, auto-merge or merge-queue entry to be pinned to that head, and skips repositories listed in `CLAUDE_MERGE_REVIEW_EXEMPT_REPOSITORIES`. Its deny reason names the review command `CLAUDE_MERGE_REVIEW_COMMAND` sets.
+- `blocking/merge_review_gate.py` denies an agent's merge until a review pass on the merged head reported zero findings, or ruled out every finding and linked the posted rulings. It reads the receipt `scripts/merge_review_receipt.py` writes, requires the merge, auto-merge or merge-queue entry to be pinned to that head, and skips repositories listed in `CLAUDE_MERGE_REVIEW_EXEMPT_REPOSITORIES`. Its deny reason names the review command `CLAUDE_MERGE_REVIEW_COMMAND` sets. The reason tells the agent to run that command for the first three passes and `/code-review low --fix <number>` for each later pass; the gate itself counts no passes.
 - `blocking/merge_over_review_offer_gate.py` denies a decision card or question that offers a merge while it names open review findings, a red check, unaddressed review comments, or an override. It reads the whole card as one text: every question, header, context, reason, and option, whether the option is a string or a label with a consequence or description.
 - `blocking/artifact_dark_mode_gate.py` denies an Artifact page publish when the page, inside the publish wrapper, keeps a light body or shows text under 4.5:1 contrast in dark mode. `blocking/artifact_dark_mode_render.cjs` renders the page in headless Chromium for it.
 

package/hooks/hooks_constants/merge_review_gate_constants.py

@@ -54,8 +54,10 @@ RECORD_COMMAND = "python3 ~/.claude/scripts/merge_review_receipt.py"
 NO_CLEAN_REVIEW_REASON_TEMPLATE = (
     "{slug}#{number} has no clean review on the head you are merging. Run the review "
     "loop: review the pull request head with `{review_command}`, fix and push every "
-    "finding, and review again "
-    "until a pass reports zero findings or every finding is ruled out with file:line "
+    "finding, and review again. After three full passes, each later pass is "
+    "`/code-review low --fix {number}`, which names the pull request so the diff "
+    "covers its base: commit and push its fixes, then review the new head. "
+    "Repeat until a pass reports zero findings or every finding is ruled out with file:line "
     "evidence. Record that pass with `"
     + RECORD_COMMAND
     + " {slug} {number} <head sha> <finding count>`, adding `--ruled-out <finding count> "

package/hooks/hooks_constants/owned_pr_review_loop_constants.py

@@ -114,7 +114,9 @@ REVIEW_LOOP_CONTEXT_TEMPLATE: str = (
     "asked for a draft or stopped at opening it.\n"
     "1. Review loop. Load the efficient-review skill (/efficient-review). Review the\n"
     "   branch, check each finding against the code, fix every one that holds, push,\n"
-    "   and review again. Repeat until a pass reports no findings.\n"
+    "   and review again. After three full passes, each later pass is\n"
+    "   /code-review low --fix {number}; commit and push its fixes, then review the new head.\n"
+    "   Repeat until a pass reports no findings.\n"
     "2. Mark ready. Once the loop is clean, CI is green, and no review thread is open,\n"
     "   mark the PR ready for review (gh pr ready {number} --repo {owner}/{repository},\n"
     "   or update_pull_request with draft false).\n"

package/package.json

@@ -1,6 +1,6 @@
 {
     "name": "claude-dev-env",
-    "version": "8.92.0",
+    "version": "8.92.1",
     "description": "Claude Code development standards — rules, hooks, agents, commands, and skills",
     "type": "module",
     "bin": {