@reviewsensei/cli-linux-x64-gnu 0.6.9 → 0.6.11

package/bin/_internal/review_sensei.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: review-sensei
-Version: 0.6.9
+Version: 0.6.11
 Summary: Provider-neutral AI code review engine with an Ollama adapter.
 Author: Mohamad Alsabbagh
 License-Expression: MIT
@@ -136,8 +136,8 @@ Node.js 22 or newer can invoke the same Python review engine through the
 provider-neutral `@reviewsensei/cli` launcher:
 
 ```bash
-npx --yes @reviewsensei/cli@0.6.9 --version
-npx --yes @reviewsensei/cli@0.6.9 --help
+npx --yes @reviewsensei/cli@0.6.11 --version
+npx --yes @reviewsensei/cli@0.6.11 --help
 ```
 
 The launcher selects one of the five native packages (macOS arm64/x64,
@@ -646,8 +646,9 @@ operator-managed `v5` public git tag. The reusable workflow prefers the
 requested exact package from PyPI and falls back to its executing workflow
 commit only when that package/version is unavailable. Unrelated PyPI
 installation failures remain fatal. Setup creates no repository variables and
-manages 3 generated files: the caller, the uninstall workflow, and the minimal
-root `.reviewsensei.yml` that the operator owns from the first install on. The
+manages 3 generated files: the caller, the uninstall workflow, and the root
+`.reviewsensei.yml` that lists every packaged default and belongs to the
+operator from the first install on. The
 only two optional Actions overrides the product consumes are
 `REVIEWSENSEI_PROVIDER` (`inference.backend`) and `REVIEWSENSEI_MODEL`
 (`inference.model`); the reusable workflow maps them once from the trusted
@@ -670,21 +671,33 @@ item only when the transport reports an oversized response. The 512 KiB
 transport ceiling remains unchanged. Fork pull requests fail closed before
 provider or broker access.
 
-When automatic review and GitHub writes are enabled, automatic approval applies
-by default; `github.reviews: blocking` or `github.reviews: advisory` in
-`.reviewsensei.yml` selects a non-approving policy instead.
-ReviewSensei publishes blocking findings as `REQUEST_CHANGES` and non-blocking
-findings as `COMMENT`, then invokes one shared, idempotent approval finalizer.
-It emits `APPROVE` for an eligible exact head when no unresolved ReviewSensei
-root is classified blocking. A later execution on the same commit still
-requests changes if blocking comments appear after an approval, and it
-approves after an earlier change request only once those blocking roots are
-resolved. Unresolved non-blocking ReviewSensei findings and human threads do
-not withhold approval. An unclassified ReviewSensei root, malformed or
-incomplete thread data, or a draft/closed/fork/stale/App-authored PR fails
-closed. The finalizer runs after review publication and after the AI resolves a
-blocking ReviewSensei thread; `@sensei` replies themselves remain ordinary
-comments. Existing approval markers prevent duplicate approvals.
+When automatic review and GitHub writes are enabled, automatic approval is on by
+default; `github.reviews: blocking` (publish and enforce, never automatically
+approve) or `github.reviews: advisory` (no ReviewSensei merge gate and no
+approval) in `.reviewsensei.yml` selects another policy.
+ReviewSensei publishes one exact-head `COMMENT` review with its summary and
+valid inline findings, and enforces through one stable `ReviewSensei` check run
+bound to that head and to the App that produced it: `success` for a complete
+review with no required fixes, `failure` when required fixes remain, and
+`action_required` for a partial, incomplete, or unpublished review. An earlier
+conclusion on the same head is superseded by a pending `in_progress` write, so
+a finished review cannot stand in for one that is still running. No persistent
+`REQUEST_CHANGES` is emitted as a second gate. A repository administrator must
+mark `ReviewSensei` required for it to gate merges; `review-sensei doctor`
+reports the check identity and producing App, and nothing reads or changes
+branch protection.
+
+The same mode also runs one shared, idempotent approval finalizer. It emits
+`APPROVE` for an eligible exact head whose review is complete and qualified and
+whose bounded thread sweep finds no unresolved blocking ReviewSensei root.
+Unresolved non-blocking ReviewSensei findings and human threads do not withhold
+approval. An unclassified ReviewSensei root, malformed or incomplete thread
+data, an unverified configuration, or a draft/closed/fork/stale/App-authored PR
+fails closed or reports why approval was withheld. The finalizer runs after
+review publication and after the AI resolves a blocking ReviewSensei thread;
+`@sensei` replies themselves remain ordinary comments. Existing approval markers
+prevent duplicate approvals, and ReviewSensei never calls a merge endpoint or
+enables auto-merge.
 
 The approval policy is deliberately conservative: explicit blocking metadata,
 or an unclassified canonical `critical`/`high` severity, blocks approval.

package/bin/_internal/review_sensei.egg-info/SOURCES.txt

@@ -77,6 +77,7 @@ docs/adr/0054-observed-convergence-acceptance-evidence.md
 docs/adr/0055-merge-focused-default-and-legacy-retirement.md
 docs/adr/0056-host-fact-placement-and-honest-approval-presentation.md
 docs/adr/0056-remove-pr-wide-review-count-caps.md
+docs/adr/0057-one-check-run-as-the-single-merge-authority.md
 docs/adr/README.md
 docs/architecture/index.md
 docs/process/adr-process.md
@@ -180,6 +181,7 @@ src/review_sensei/hosting/github/approval.py
 src/review_sensei/hosting/github/auth.py
 src/review_sensei/hosting/github/broker.py
 src/review_sensei/hosting/github/broker_client.py
+src/review_sensei/hosting/github/checks.py
 src/review_sensei/hosting/github/conversation.py
 src/review_sensei/hosting/github/errors.py
 src/review_sensei/hosting/github/http.py

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@reviewsensei/cli-linux-x64-gnu",
-  "version": "0.6.9",
+  "version": "0.6.11",
   "description": "ReviewSensei native CLI for glibc Linux x64",
   "license": "MIT",
   "repository": {