@reviewsensei/cli-win32-x64 0.1.1 → 0.6.6

package/bin/_internal/{review_sensei-0.1.1.dist-info → review_sensei-0.6.6.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: review-sensei
-Version: 0.1.1
+Version: 0.6.6
 Summary: Provider-neutral AI code review engine with an Ollama adapter.
 Author: Mohamad Alsabbagh
 License-Expression: MIT
@@ -36,11 +36,14 @@ providers later.
 
 > Alpha: this repository provides an open-source review engine and CLI. The
 > primary usage path is running ReviewSensei in your own GitHub Actions
-> workflow. Setup-v4 callers follow the operator-managed public `v4` tag; the
+> workflow. Setup-v4 callers follow the operator-managed public `v5` tag; the
 > Worker validates that tag before creating the setup PR and the broker resolves
-> it again when issuing a capability. The example workflow installs the
+> it again when issuing a capability. The broker still accepts `v4` during
+> migration. The example workflow installs the
 > requested exact package from PyPI first and falls back to its executing commit
-> only when the distribution is unavailable. It
+> only when the distribution is unavailable. Both paths prove the intended
+> release identity against a compatibility manifest; live canary and `v5`
+> movement remain operator-only. It
 > defaults to a local Ollama endpoint and does not require a hosted backend.
 
 ## What it does
@@ -48,7 +51,10 @@ providers later.
 - Keeps review business logic independent of GitHub and model vendors.
 - Sends a structured prompt to a replaceable provider adapter.
 - Requires provider output to be valid JSON.
-- Rejects inline comments that target files or lines outside the supplied diff.
+- Filters inline comments that target files or lines outside the supplied diff
+  while preserving independently valid findings, including deletions and
+  file-level concerns, and reports an explicit coverage outcome for every
+  changed file.
 - Supports local Ollama servers and Ollama Cloud through the same adapter.
 - Loads approved repository-local learnings and returns validated learning proposals.
 - Supports ordered review stages with structured categories and explicit focus
@@ -59,9 +65,13 @@ providers later.
   runs and one provider slot per pull request.
 - Publishes versioned JSON Schemas for review results, learnings, categories,
   stages, and concurrency plans.
+- Adds optional blocking, severity, fix-effort, and lens labels with
+  deterministic summary counts and visual cues for fast triage.
 - Provides a narrow GitHub App JWT and installation-token auth adapter for
   optional App-identity comments or reviews.
 - Avoids retaining raw prompts and provider responses in the engine.
+- Provides `review-sensei doctor` and a zero-write `review-sensei plan`
+  preview so skipped or failed setup is diagnosable without forking workflows.
 
 ## Quick start
 
@@ -88,7 +98,7 @@ Ollama Cloud is explicit opt-in:
 ```bash
 export OLLAMA_API_KEY="your-key"
 export REVIEWSENSEI_PROVIDER_MODE="cloud"
-export REVIEWSENSEI_CLOUD_MODEL="deepseek-v4-flash:cloud"
+export REVIEWSENSEI_CLOUD_MODEL="deepseek-v4.1-flash:cloud"
 unset OLLAMA_BASE_URL OLLAMA_MODEL
 ```
 
@@ -98,8 +108,8 @@ Node.js 22 or newer can invoke the same Python review engine through the
 provider-neutral `@reviewsensei/cli` launcher:
 
 ```bash
-npx --yes @reviewsensei/cli@0.1.1 --version
-npx --yes @reviewsensei/cli@0.1.1 --help
+npx --yes @reviewsensei/cli@0.6.6 --version
+npx --yes @reviewsensei/cli@0.6.6 --help
 ```
 
 The launcher selects one of the five native packages (macOS arm64/x64,
@@ -113,7 +123,7 @@ The package version is intentionally identical to the Python project and
 release tag. See [`docs/installation.md`](docs/installation.md),
 [`docs/releasing.md`](docs/releasing.md), and [ADR 0031](docs/adr/0031-npm-launcher-and-standalone-platform-packages.md)
 for prerequisites, release ordering, provenance, and version-forward rollback
-guidance (internal tracking issue #103).
+guidance.
 
 Generate a bounded diff and run a review:
 
@@ -123,6 +133,8 @@ review-sensei prepare-diff \
   --base-ref main \
   --head-ref feature \
   --output pr.patch
+review-sensei doctor
+review-sensei plan --diff pr.patch --repository owner/repository --pull-request 42
 review-sensei \
   --diff pr.patch \
   --repository owner/repository \
@@ -156,30 +168,48 @@ Optional App-identity publication is documented in
 authentication adapter for GitHub App JWTs and installation access tokens; it
 is not a hosted review service, webhook receiver, or durable delivery store.
 
-### GitHub App setup-v4 integration
+### GitHub App setup-v5 integration
 
-The optional setup-v4 integration adds a reviewable, generated caller by using
-the operator-managed `v4` git tag as the only update channel:
-`malsabbagh/review-sensei/.github/workflows/review-sensei-run.yml@v4`.
+The optional setup-v5 integration adds a reviewable, generated caller by using
+the operator-managed `v5` git tag as the only update channel:
+`malsabbagh/review-sensei/.github/workflows/review-sensei-run.yml@v5`.
 The Worker validates the tag before writing the caller, and the Cloudflare
 broker resolves the same tag at capability exchange time and checks the
-executing workflow SHA. Moving `v4` is therefore the public cutoff action.
+executing workflow SHA. Moving `v5` is therefore the public cutoff action.
 The selected provider mode applies consistently to automatic pull-request
 reviews, manual reviews, learning proposals, artifacts, and authorized
 `@sensei` conversations. Cloud runs use GitHub-hosted compute; local runs use
 the operator-controlled labeled self-hosted runner. Every generated write
-switch defaults to `false`.
+switch except automatic approval defaults to `false`; automatic approval
+defaults to `true` and requires automatic review and GitHub writes as separate
+gates.
 
 Cloud mode may pass the existing customer-owned `OLLAMA_API_KEY` secret by
 name (`secrets.OLLAMA_API_KEY`) to the public reusable workflow. The App
 does not create, read, persist, log, or reveal that secret value. The Worker
-broker only issues one capability-scoped installation token after verifying
-the signed Actions OIDC identity, exact v4 workflow ref and runtime SHA,
+broker issues a short-lived, capability-scoped installation token for each
+authorized exchange after verifying
+the signed Actions OIDC identity, exact `v5` workflow ref and runtime SHA,
 repository identity, fork status, installation, replay state, and rate limit.
+For each write operation, the workflow obtains a separate short-lived token for
+the requested capability: `review_publish` for review publication,
+`inline_reply` or `issue_reply` for replies, and `learning_write` for learning
+pull requests. The broker-issued token grants the workflow the capability to
+make the corresponding GitHub API write; Cloudflare does not perform that
+write. Review and reply capabilities grant `pull_requests: write`;
+`learning_write` additionally grants `contents: write`. A review-publication
+token cannot write repository contents.
+Cloudflare does not run the review engine or send prompts to the model provider;
+the GitHub Actions job performs those operations in the configured provider
+environment.
+
+During installation or migration, the Worker separately uses GitHub App
+authentication to obtain a repository-scoped installation token for the setup
+branch, pull request, and default variables.
 
 Installation, reinstall, permission-acceptance, and repository-added events
-reconcile absent or older generated clients through at most one setup-v4 PR.
-Current v4 is a no-op; managed v3 and older generated clients are migrated
+reconcile absent or older generated clients through at most one setup-v5 PR.
+Current tag-following setup-v5 is a no-op; managed v3/v4 and older generated clients are migrated
 through a setup PR, while custom, malformed, and future setup files are no-write
 cases. See [`docs/installation.md`](docs/installation.md),
 [`docs/github-app-registration.md`](docs/github-app-registration.md), and the
@@ -189,8 +219,8 @@ that never distributes the App private key.
 
 ## Offline evaluation
 
-Issue #9 includes a synthetic, CC0-1.0 corpus that runs without network access
-or provider credentials:
+The repository includes a synthetic, CC0-1.0 corpus that runs without network
+access or provider credentials:
 
 ```bash
 python scripts/validate_evaluation_corpus.py
@@ -233,6 +263,15 @@ bytes for the compact UTF-8 serialization of the complete proposal array
 result; and line numbers up to 2,147,483,647. These are byte limits unless a
 unit is stated otherwise.
 
+Every changed file in a review result receives an explicit coverage outcome
+(`reviewed`, `partially-reviewed`, `excluded-by-policy`, `unsupported`, or
+`budget-exhausted`). Incomplete enumeration is never reported as fully
+reviewed. Finding locations may bind to a right-side added line (the legacy
+default), a left-side deleted line, or a file-level concern. Per-request
+limits stay in place; opt-in `--orchestrate-large-changes` adds a separate
+total-work budget so larger diffs can be chunked without silent truncation or
+unbounded provider calls.
+
 Repository paths are strict canonical NFC UTF-8, `/`-separated, relative paths:
 no empty, dot, parent, absolute, drive/UNC, backslash, surrounding whitespace,
 control/format/surrogate, or silently normalized forms are accepted. Glob
@@ -240,7 +279,10 @@ tokens are allowed only in pattern fields; real Git paths may contain literal
 glob characters and are preserved without being interpreted. Git C-quoted
 paths are decoded as strict bytes before the same path checks. Exact duplicate
 comments identified by `(path, line, body)` use stable first-wins ordering;
-comments that differ in body remain distinct. Expected failures are sanitized
+comments that differ in body remain distinct. Across heads, findings also carry
+a stable fingerprint from path, symbol, defect kind, and optional evidence id
+so a moved or rephrased concern does not open a duplicate ReviewSensei thread,
+and a later pass that omits a finding is not treated as a fix. Expected failures are sanitized
 and never include a prompt, diff, provider body, credential, or supplied secret
 marker. Invalid structured provider output receives one fresh, sanitized
 correction attempt per stage; the rejected attempt is discarded completely and
@@ -306,17 +348,21 @@ category to a comment, it must use one of the active stage ids; classification
 is optional for compatibility. If multiple stages reuse an id, the entire lens
 definition must be identical.
 
-Inline findings may also carry independent `severity` (`critical`, `high`,
-`medium`, or `low`) and `fix_effort` (`trivial`, `small`, `moderate`, `large`,
-or `unknown`) labels. Severity describes likely impact, while fix effort
-describes remediation scope rather than a time estimate. The existing
-`category` id is the lens source and is shown as a human-readable lens label;
-no composite priority score or second `lens` field is emitted. These fields are
-optional so legacy provider results remain valid. When present, the same labels
-appear on inline GitHub comments and deterministic severity/lens/quick-win
-counts appear in the review summary. Labels are validated as printable,
-bounded provider text and Markdown-escaped before publication so untrusted
-output cannot inject structure or exceed the configured comment-body limit.
+Inline findings may also carry independent `blocking`, `severity`
+(`critical`, `high`, `medium`, or `low`), and `fix_effort` (`trivial`, `small`,
+`moderate`, `large`, or `unknown`) labels. `blocking` controls whether the
+finding prevents approval; `false` identifies an optional follow-up.
+If it is omitted, only case-insensitive canonical `critical` and `high`
+severity values block approval. Missing, lower-severity, and legacy free-form
+severity values are non-blocking.
+Severity describes likely impact, while fix effort describes remediation scope
+rather than a time estimate. The existing `category` id is the lens source and
+is shown as a human-readable lens label; no composite priority score or second
+`lens` field is emitted. When present, the same labels appear on inline GitHub
+comments and deterministic blocking/severity/lens/quick-win counts appear in
+the review summary. Labels are validated as printable, bounded provider text
+and Markdown-escaped before publication so untrusted output cannot inject
+structure or exceed the configured comment-body limit.
 The formatted summary is checked against `max_summary_bytes`, and the complete
 framed GitHub review body is bounded to 65,536 bytes before publication.
 
@@ -378,6 +424,47 @@ packaged default pipeline includes an architecture lens whose optional sources
 are `AGENTS.md`, `docs/architecture.md`, `docs/architecture/**/*.md`, and
 `docs/adr/**/*.md`.
 
+### Opt-in symbol-aware source context
+
+By default, reviews do **not** load related source files. That keeps the
+trusted context boundary at configured documents and approved learnings.
+Operators may opt in to bounded Python symbol-aware context from the same
+trusted base checkout:
+
+```bash
+review-sensei --diff pr.patch \
+  --context-root /path/to/target-branch-checkout \
+  --enable-symbol-context \
+  --base-sha <trusted-base-commit-sha>
+```
+
+`--enable-symbol-context` can also be set with
+`REVIEWSENSEI_ENABLE_SYMBOL_CONTEXT=true`. `--base-sha` is required when the
+selector is enabled and identifies the immutable trusted-base snapshot. Optional
+`--head-sha` is recorded only as untrusted coverage metadata; head trees, build
+scripts, hooks, tests, and package installers are never checked out or
+executed for this context.
+
+The documented language set is Python. The selector uses static AST parsing
+for enclosing functions/classes, local imports, Protocol/ABC interfaces,
+same-directory callers, and conventional associated tests. JavaScript and
+TypeScript files may be included as a whole-file fallback and are reported as
+`unsupported-language`. Canonical path, traversal, symlink, file-type, and
+secret-sensitive-file protections apply. Identical snapshot and policy inputs
+produce identical excerpts and outcome ordering.
+
+Each excerpt records path, base commit SHA, Git blob object id, line range,
+selection reason, and SHA-256 digest. Ambiguous parses, excluded paths,
+unsupported languages, and exhausted budgets appear in result
+`source_context` coverage; they are not presented as complete understanding.
+Bounded relationship expansion names the bound it reached, using
+`directory-truncated` for a capped caller-directory listing and
+`relations-truncated` for a capped import or caller candidate set.
+When the opt-in policy is enabled, exhausting the configured file or byte
+budget fails closed. Evaluation under
+[#33](https://github.com/malsabbagh/review-sensei/issues/33) is required
+before this selector can become the default.
+
 ## Repository-local learnings
 
 Durable repository knowledge lives in the reviewed repository under
@@ -403,6 +490,33 @@ must be a checkout of the target branch or target commit, not the untrusted
 head checkout. This prevents a pull request from changing its own review
 context before it is merged.
 
+Existing files may omit lifecycle metadata. Optional bounded fields include
+`owner`, `provenance`, `reviewed_at`, `expires_at`, `supersedes`, and
+`superseded_by`. Status may be `active`, `retired`, or `superseded`.
+Superseded and retired entries stay on disk for provenance and Git rollback,
+but they are not selected for later prompts. Diagnostics for stale,
+conflicting, missing-superseder, or cyclic entries are advisory; ReviewSensei
+never deletes or rewrites approved knowledge automatically.
+
+```bash
+review-sensei learnings diagnose --learning-root /path/to/target-branch --json
+```
+
+Opt-in finding feedback is a separate `learning-feedback` document with
+outcomes `useful`, `incorrect`, `obsolete`, and `unverified`. Absence of
+feedback is not approval, and feedback never becomes trusted review context.
+
+Fixture evaluation can compare the same cases with and without selected
+approved learnings. The report records precision/recall/false-positive deltas
+as estimates, not causal proof:
+
+```bash
+review-sensei evaluate --mode fixture --corpus evaluation/v1/corpus.json \
+  --compare-learnings --learning-root /path/to/target-branch
+```
+
+See [ADR 0036](docs/adr/0036-learning-lifecycle-diagnostics-and-feedback.md).
+
 ## Local Ollama
 
 ```bash
@@ -414,6 +528,12 @@ review-sensei --diff pr.patch
 
 ## Providers
 
+The shipped registry keys are `ollama`, `openai-compatible`, and `openrouter`.
+Operator defaults, workflow modes, named profiles, credentials, and hosted
+OpenRouter allowlists are documented in [`docs/installation.md`](docs/installation.md).
+The release-aware public matrix lives at
+[reviewsensei.dev/providers/](https://reviewsensei.dev/providers/).
+
 The core depends on this small interface:
 
 ```python
@@ -429,19 +549,30 @@ provider's API and returning `ProviderResponse`. The review service, diff
 validation, output schema, and GitHub-independent behavior remain unchanged.
 See [the architecture guide](docs/architecture.md).
 
+The OpenAI-compatible adapter is restricted to `https://api.openai.com` by
+default and rejects redirects so its bearer token cannot be silently forwarded
+to another host. An explicitly trusted private or compatible service may be
+selected from Python with `allow_custom_endpoint=True`; this is an operator
+acknowledgement of data egress and still requires a non-empty API key. Custom
+openers own their transport and redirect policy, while the built-in opener uses
+the system CA store (or an explicitly configured, regular `SSL_CERT_FILE`) and
+reuses one verified TLS context for the provider lifetime. Local Ollama and
+Ollama Cloud remain separate provider profiles; selecting Cloud is the explicit
+network egress path for Ollama deployments.
+
 ## GitHub integration
 
-The GitHub App opens a setup-v4 PR containing a thin caller pinned to the full
-SHA resolved from the operator-managed `v4` public git tag at installation
-time. The reusable workflow prefers the requested exact package from PyPI and,
-only when that package/version is unavailable, installs the executing workflow
-commit directly from the public GitHub source. Unrelated PyPI installation
-failures remain fatal. All nine
-`REVIEWSENSEI_*` repository variables are created with safe defaults: automatic
-review, GitHub writes, learning PRs, mention replies, and artifact upload are
-off. The App never creates the customer-owned
-`OLLAMA_API_KEY` secret.
-The five feature switches are all disabled by default.
+The GitHub App opens a setup-v5 PR containing a thin caller that follows the
+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 provisions 15 repository variables,
+including 7 boolean controls, and manages 3 generated files. Automatic review,
+GitHub writes, learning proposals, learning PRs, mention replies, and artifact
+upload default to `false`; automatic approval defaults to `true` but is effective
+only when review and write gates are enabled. `REVIEWSENSEI_REVIEW_MODE` is a
+policy selector, not a boolean control, and defaults to `merge-focused`.
+The App never creates the customer-owned `OLLAMA_API_KEY` secret.
 
 When explicitly enabled, same-repository pull requests can run automatic review
 with either provider mode: cloud on a GitHub-hosted runner or local Ollama on
@@ -455,19 +586,27 @@ 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, ReviewSensei selects
-`APPROVE` for a validated exact-head result only when it has no inline findings
-and a final bounded GitHub thread sweep confirms that every existing review
-thread is resolved. Any open finding or thread, draft/closed/fork/stale PR,
-App-authored PR, or incomplete thread lookup keeps the event as `COMMENT` or
-fails closed; `@sensei` replies are always ordinary comments. Existing review
-markers still deduplicate one publication per head, so resolving a thread does
-not retroactively approve an already-published review.
-
-The approval policy is deliberately conservative: severity, fix effort, lens,
-and model confidence help triage findings but never override an unresolved
-thread. A later head must receive a fresh clean review after the final
-resolution sweep. See [ADR 0030](docs/adr/0030-gate-app-approvals-on-resolved-review-threads-and-exact-head-review-safety.md)
+When automatic review and GitHub writes are enabled, automatic approval is on by
+default; set `REVIEWSENSEI_AUTO_APPROVE=false` for an explicit opt-out.
+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.
+
+The approval policy is deliberately conservative: explicit blocking metadata,
+or an unclassified canonical `critical`/`high` severity, blocks approval.
+Resolving a thread manually does not start GitHub Actions, so rerun the review
+workflow (or push a new head) if a manually resolved blocking root was the last
+blocker. See
+[ADR 0032](docs/adr/0032-blocking-finding-classification-for-approvals.md)
 for the criteria and rollback path.
 
 Manual review and authorized `@sensei` replies use the same selected provider
@@ -479,10 +618,10 @@ mode sends that bounded conversation context to Ollama Cloud; local mode keeps
 it on the configured local service. Published review summaries and inline
 findings also tell readers to reply with @sensei followed by their question.
 `review.json` is uploaded only when
-`REVIEWSENSEI_UPLOAD_ARTIFACTS=true`; setup-v4 does not create a separate
+`REVIEWSENSEI_UPLOAD_ARTIFACTS=true`; setup-v5 does not create a separate
 version artifact. See [installation and migration](docs/installation.md) and
 the reviewable
-[`setup-v4 example`](examples/github-actions/review-sensei-review.yml).
+[`setup-v5 example`](examples/github-actions/review-sensei-review.yml).
 
 ## Concurrency policy
 
@@ -490,11 +629,14 @@ Hosts that run more than one review worker can reuse the same deterministic
 policy:
 
 ```python
-from review_sensei import ReviewConcurrencyPlan, ReviewService
+from review_sensei import ProviderAdmission, ReviewConcurrencyPlan, ReviewService
 
 plan = ReviewService(provider).concurrency_plan(request)
 # Enforce plan.workflow in the run scheduler.
 # Enforce plan.provider in the provider-stage scheduler.
+admission = ProviderAdmission(plan.provider)
+with admission.acquire():
+    result = ReviewService(provider).review(request)
 ```
 
 The workflow group is keyed by repository and pull request and is latest-wins.
@@ -502,8 +644,14 @@ The provider group uses the same pull-request identity, admits one active
 provider stage, and does not cancel the active stage. Different pull requests
 therefore remain independent. `ReviewConcurrencyPlan.for_non_review_trigger()`
 creates an isolated workflow group with no provider slot for triggers that do
-not request a review. ReviewSensei returns this policy but does not own durable
-queue state or cross-process cancellation.
+not request a review.
+
+GitHub-hosted runs enforce this with native workflow `concurrency` groups
+(`max_active=1`, cancel-in-progress for reviews). Local hosts and tests can
+use `ProviderAdmission` to enforce the same bound in-process, including lease
+release after cancellation or failure and a hard cap on waiters. ReviewSensei
+does not own durable queue state, a global cross-repository lock, or
+cross-process cancellation.
 
 ## Data handling
 
@@ -540,4 +688,10 @@ See [CONTRIBUTING.md](CONTRIBUTING.md), [SECURITY.md](SECURITY.md), and
 
 ## License
 
-ReviewSensei is released under the [MIT License](LICENSE).
+ReviewSensei is released under the [MIT License](LICENSE). The synthetic
+evaluation corpus under `evaluation/v1/` is CC0-1.0. For the ownership
+inventory, trademark usage, contribution licensing (DCO; no CLA), and
+Community / Managed / Enterprise boundary, see
+[`docs/ownership-and-licensing.md`](docs/ownership-and-licensing.md),
+[`TRADEMARKS.md`](TRADEMARKS.md), [`CONTRIBUTING.md`](CONTRIBUTING.md), and
+[ADR 0040](docs/adr/0040-community-managed-enterprise-boundary.md).

package/bin/_internal/review_sensei-0.6.6.dist-info/RECORD

@@ -0,0 +1,175 @@
+../../Scripts/review-sensei.exe,sha256=aiJfPftvZjcLCqupusFwyry5-u0omCFFlhRXj-FZHUM,108355
+review_sensei-0.6.6.dist-info/INSTALLER,sha256=zuuue4knoyJ-UwPPXg8fezS7VCrXJQrAP7zeNuwvFQg,4
+review_sensei-0.6.6.dist-info/METADATA,sha256=nHIIFWJW9glRh0Yd0BgE0yJYfE8EP12hZ52kGDsaMLM,34155
+review_sensei-0.6.6.dist-info/RECORD,,
+review_sensei-0.6.6.dist-info/REQUESTED,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+review_sensei-0.6.6.dist-info/WHEEL,sha256=YVMoNqKzERt-wjUZwJ33xBGAwnFl-4cqbYkTtWa4itE,91
+review_sensei-0.6.6.dist-info/direct_url.json,sha256=EDhpXdSFq1qZfl2FCT07Hg2gleInjGFaAwkFy0selqU,67
+review_sensei-0.6.6.dist-info/entry_points.txt,sha256=Eki1ToC6hiOv8PMEAd2Cnbh5WWstVzzzf0XmHBHzpMQ,57
+review_sensei-0.6.6.dist-info/licenses/LICENSE,sha256=-eVBqOV92YK1K76UIcK5MrgUh0EiQMe8xNhSysQE68s,1095
+review_sensei-0.6.6.dist-info/top_level.txt,sha256=VCF4gJIJPP3MPJntoITDjMhJ8MUZfVLj-z3-VUQ9XDY,14
+review_sensei/__init__.py,sha256=hBA7OkQbK_oV4-fDFucVnpsnIut_JK-BbXLTznhg8ao,10524
+review_sensei/__main__.py,sha256=0iDYjdmiXNBinFF23WUyotsdpA6wFKjoB5aw3esTSRg,83
+review_sensei/__pycache__/__init__.cpython-311.pyc,,
+review_sensei/__pycache__/__main__.cpython-311.pyc,,
+review_sensei/__pycache__/baseline.cpython-311.pyc,,
+review_sensei/__pycache__/cli.cpython-311.pyc,,
+review_sensei/__pycache__/concurrency.cpython-311.pyc,,
+review_sensei/__pycache__/context.cpython-311.pyc,,
+review_sensei/__pycache__/convergence.cpython-311.pyc,,
+review_sensei/__pycache__/conversation.cpython-311.pyc,,
+review_sensei/__pycache__/coverage.cpython-311.pyc,,
+review_sensei/__pycache__/diagnostics.cpython-311.pyc,,
+review_sensei/__pycache__/diff.cpython-311.pyc,,
+review_sensei/__pycache__/disposition.cpython-311.pyc,,
+review_sensei/__pycache__/errors.cpython-311.pyc,,
+review_sensei/__pycache__/evaluation.cpython-311.pyc,,
+review_sensei/__pycache__/learnings.cpython-311.pyc,,
+review_sensei/__pycache__/models.cpython-311.pyc,,
+review_sensei/__pycache__/openrouter_qualification.cpython-311.pyc,,
+review_sensei/__pycache__/outcomes.cpython-311.pyc,,
+review_sensei/__pycache__/patches.cpython-311.pyc,,
+review_sensei/__pycache__/planning.cpython-311.pyc,,
+review_sensei/__pycache__/presentation.cpython-311.pyc,,
+review_sensei/__pycache__/provider_config.cpython-311.pyc,,
+review_sensei/__pycache__/release_manifest.cpython-311.pyc,,
+review_sensei/__pycache__/schemas.cpython-311.pyc,,
+review_sensei/__pycache__/sequence.cpython-311.pyc,,
+review_sensei/__pycache__/service.cpython-311.pyc,,
+review_sensei/__pycache__/session.cpython-311.pyc,,
+review_sensei/__pycache__/stages.cpython-311.pyc,,
+review_sensei/__pycache__/validation.cpython-311.pyc,,
+review_sensei/__pycache__/verifier.cpython-311.pyc,,
+review_sensei/__pycache__/workflow.cpython-311.pyc,,
+review_sensei/baseline.py,sha256=nJegFpLDW-YviQqUt8QXWSOgzuPIWg3ryz1DTr87d4U,54311
+review_sensei/cli.py,sha256=CarOAUk2BhP2vsIzmMq6IPtJV5gc2d78Z0rqD0ZJjm8,130843
+review_sensei/concurrency.py,sha256=7nqMDNaIl7YoKafkAF1ctXu6D5SFxXosm-1407OKimQ,20097
+review_sensei/context.py,sha256=ejDs0EBaUVTbAkrs-wACGXbgQcEwClX0Tultn2Mo_R0,96255
+review_sensei/convergence.py,sha256=XJQ3Fx4N41Fzf_Y2PVeebyS3tf7e7cpbD1p4VdPuHTg,54234
+review_sensei/conversation.py,sha256=KRWtF8JnmOrOQsvAX4q-d0UzzXjNp2XeePSbyIm4_pw,7467
+review_sensei/coverage.py,sha256=8fVsO4x2swqaa9QI5fcqeqB_FB6m0JfKrlRCg7_KA6E,12951
+review_sensei/default_categories/01-correctness.json,sha256=ZeTTvEavHU6_ih7JKhv8a6gH-4wRJGv0abckRWNWTfs,211
+review_sensei/default_categories/02-security.json,sha256=6mz8Hlteb9q1PouLDmDoR-1wxrN31XR2X5I9yoYiG-E,231
+review_sensei/default_categories/03-architecture.json,sha256=Fg3VsUCzl9trCpSnoOgwTqfNHbB7a4znEfrw2apqjXA,806
+review_sensei/default_categories/03-maintainability.json,sha256=eyF9_slQ50LwEqSSet63jthfNp9ZOTnwiDVtVYjVv1A,229
+review_sensei/default_categories/04-tests.json,sha256=UHhvH79bae9FmLJT_MYNvki4Y9xhJDLpDuLdgp2npe0,208
+review_sensei/default_stages/01-default-review.json,sha256=-y9LP_8XNqEPpRN2K8z_qMHRqXyP6ZsiydzZFHozZwY,3463
+review_sensei/diagnostics.py,sha256=CAYZPlOBlJ8ZNyPUrbzX-bcnbigxhA593_aaGqNj5hY,44971
+review_sensei/diff.py,sha256=na5MWhYcnRimFwGElNBiSp27Su_oufHhWiGtm1W38oY,32057
+review_sensei/disposition.py,sha256=jENlqe7JU3kPCBo-0_jxpfEkb_rltcEvXoRihstmo7s,24908
+review_sensei/errors.py,sha256=1-JKbLcEodjQMXleJwR-QDzRhMSFwDSe9Oc6Pg1UYmY,2901
+review_sensei/evaluation.py,sha256=kKhbcF3XpHhDYLW8e_dxGjcS0sXJpz8nXYsPpUuXnBE,59450
+review_sensei/hosting/__init__.py,sha256=zqIukHwfUzai3x7dJS2nFNlVhP7EQWQ47-4pGCi2Sz0,84
+review_sensei/hosting/__pycache__/__init__.cpython-311.pyc,,
+review_sensei/hosting/__pycache__/openrouter_workflow.cpython-311.pyc,,
+review_sensei/hosting/github/__init__.py,sha256=x-mVioarkGsUF8Uwh4SfpEMOu51lP9p6tA4N0G6pB8o,5452
+review_sensei/hosting/github/__pycache__/__init__.cpython-311.pyc,,
+review_sensei/hosting/github/__pycache__/application.cpython-311.pyc,,
+review_sensei/hosting/github/__pycache__/approval.cpython-311.pyc,,
+review_sensei/hosting/github/__pycache__/auth.cpython-311.pyc,,
+review_sensei/hosting/github/__pycache__/broker.cpython-311.pyc,,
+review_sensei/hosting/github/__pycache__/broker_client.cpython-311.pyc,,
+review_sensei/hosting/github/__pycache__/conversation.cpython-311.pyc,,
+review_sensei/hosting/github/__pycache__/errors.cpython-311.pyc,,
+review_sensei/hosting/github/__pycache__/http.cpython-311.pyc,,
+review_sensei/hosting/github/__pycache__/jwt.cpython-311.pyc,,
+review_sensei/hosting/github/__pycache__/learning_pr.cpython-311.pyc,,
+review_sensei/hosting/github/__pycache__/observed.cpython-311.pyc,,
+review_sensei/hosting/github/__pycache__/oidc.cpython-311.pyc,,
+review_sensei/hosting/github/__pycache__/publication.cpython-311.pyc,,
+review_sensei/hosting/github/__pycache__/session_ledger.cpython-311.pyc,,
+review_sensei/hosting/github/__pycache__/setup.cpython-311.pyc,,
+review_sensei/hosting/github/__pycache__/trigger.cpython-311.pyc,,
+review_sensei/hosting/github/__pycache__/webhooks.cpython-311.pyc,,
+review_sensei/hosting/github/application.py,sha256=L9GyMXzvtxZkFXaG9XWWitf2Z974bV9krwp8VVxCb0k,70460
+review_sensei/hosting/github/approval.py,sha256=nx8h8oDvljrdrXkNahMXrvk3kMVkMy6owztXZPRwsqA,4189
+review_sensei/hosting/github/auth.py,sha256=koITagUX9piVboGJ_smEhPsCf4rOQkOqqSKeLKvFRbc,19055
+review_sensei/hosting/github/broker.py,sha256=FFC7DsFhzy1tENXTCmLNuYzX2cHaYsgMkDGwXEcJaTU,18014
+review_sensei/hosting/github/broker_client.py,sha256=v8_PQe7yF0kXtZOEbXxWcw2xW-Z7COQt9Zi8CTsQkHM,16573
+review_sensei/hosting/github/conversation.py,sha256=j0WNXWEcLLjaAHhYkk0WTbLHUXyikDKjhppCaEgImLY,62918
+review_sensei/hosting/github/errors.py,sha256=K5Lfsh4uzuAW6tER2l7pcEM4Lz6FwGEmMPxvF1vNc4E,5352
+review_sensei/hosting/github/fixtures/__init__.py,sha256=2sBygkvppFJ6F8KV1_MDmbBNl7Yd0oPvYOFG6R_NOpA,82
+review_sensei/hosting/github/fixtures/__pycache__/__init__.cpython-311.pyc,,
+review_sensei/hosting/github/fixtures/historical-v4-uninstall.yml,sha256=VrX1XpTrOlqscHDoJM_yRKwbP-cvcFvnG79G3HTO3hY,2664
+review_sensei/hosting/github/fixtures/merge-focused-v4-caller.yml,sha256=EGs0F3O7Ictx0azK07q_6Px2cMTA87mKCLnsWSyIoFY,17496
+review_sensei/hosting/github/fixtures/merge-focused-v4-config.yml,sha256=4L1UvSucMGJqPh_OwmPgq_lE-nenySkIg7Xogtmk7Us,464
+review_sensei/hosting/github/fixtures/released-v4-resolve-trigger-runner-switch.yml,sha256=09um6w5QmIodLMrbR10z4gdZ3f5AaVj7eqjhQzOfkV8,16194
+review_sensei/hosting/github/http.py,sha256=6Dz9S-U4M07KehE7FhjNyrBt_XKKz4Dnp6zULMFKHk0,8943
+review_sensei/hosting/github/jwt.py,sha256=mTakXBxU-GR-hBnIZE-FryfFXwpS9hJ7qb3X7P0FBMo,4582
+review_sensei/hosting/github/learning_pr.py,sha256=hT7FmmpWCnUO5WCQEmP9XioQiceNALxuJCtcZXAQxSU,112424
+review_sensei/hosting/github/observed.py,sha256=Pin563D-wNQVeC0_PSrh8W1138K5xM2jbN_lE2MyK6s,36230
+review_sensei/hosting/github/oidc.py,sha256=h3UVcBYQIKy-xQS7Ixv3_fy670mmuBoNKSYurPZWZ60,13310
+review_sensei/hosting/github/publication.py,sha256=h-j_FpqftFwC7gjO-jvrtUwayL31LKISECC7AGJxDTQ,85760
+review_sensei/hosting/github/session_ledger.py,sha256=bqalhVAz6AolacYn-paxdddofl7uyeR_kZE7s750iqo,31057
+review_sensei/hosting/github/setup.py,sha256=jhMoMbnlZBO2vPb1X8kdYnjC_ArHl1EMENR70J7N15A,107713
+review_sensei/hosting/github/trigger.py,sha256=ewwnF02jV1llat5xXwXM3AzZLpMLoD3El0gLKIhsIiM,11395
+review_sensei/hosting/github/webhooks.py,sha256=7hh_YlF7riV6wov7A_27aOb5g6-Ivob1PBWR4m5bF3I,13662
+review_sensei/hosting/openrouter_workflow.py,sha256=qepVpa12sL6mTaBqTGhcWEaYI8zawcgAYK6sOCRnKtE,1330
+review_sensei/learnings.py,sha256=AF9Dqyreb9WQLWLZiciuyRXNviSso6sYVTe2xzx1ZfY,21434
+review_sensei/models.py,sha256=UL3bnferZPU6ra5ris6JDJ0XMIjgvMqbgCYM8YtmeU8,93017
+review_sensei/openrouter_qualification.py,sha256=UKUL8oK8oexX1P_othR3y_QAAW0PG2EdAYnWXA-ACZI,30296
+review_sensei/outcomes.py,sha256=uFpb5-n3mvuQ71ljGQp7KHAmjtCHx13YbiXnTkmCiuk,32183
+review_sensei/patches.py,sha256=P9plsU7-5Z-Jxy8EJBFQX5hX3bcmxBoRJxTnEqA_tbA,15981
+review_sensei/planning.py,sha256=6rb6w-tNhXCBOl6-_N6A2fQox6cGjL88kOAhn4oGTpA,23289
+review_sensei/presentation.py,sha256=JlbvtcFV_Kok_21ATtDNj-kHfcmCpk5v7N15RulAAVQ,6587
+review_sensei/provider_config.py,sha256=Hgs0oNI6sNYJxuMi38dZ8YyF-Nu1i5CF7mi0K_2Kr0Q,16385
+review_sensei/providers/__init__.py,sha256=gmfB6gm0QN_FfN3lJJ0kN8QgTtJom4ayzaqUkLmwQwc,779
+review_sensei/providers/__pycache__/__init__.cpython-311.pyc,,
+review_sensei/providers/__pycache__/base.cpython-311.pyc,,
+review_sensei/providers/__pycache__/fixture.cpython-311.pyc,,
+review_sensei/providers/__pycache__/ollama.cpython-311.pyc,,
+review_sensei/providers/__pycache__/openai_compatible.cpython-311.pyc,,
+review_sensei/providers/__pycache__/openrouter.cpython-311.pyc,,
+review_sensei/providers/__pycache__/profiles.cpython-311.pyc,,
+review_sensei/providers/__pycache__/registry.cpython-311.pyc,,
+review_sensei/providers/__pycache__/routing.cpython-311.pyc,,
+review_sensei/providers/__pycache__/transport.cpython-311.pyc,,
+review_sensei/providers/base.py,sha256=u-z9bQNi3sjPudoDb52czjOmM2YwBX_Kwniopz_LfWI,2009
+review_sensei/providers/fixture.py,sha256=-HXetG6cXpWd5NlF1-RMA3aqyr88YHwYpiyz0Bc6BO8,1554
+review_sensei/providers/ollama.py,sha256=50u86LU-FYulbBxJP_vShGw-9t2sGCVfceGE8vjZdFY,10739
+review_sensei/providers/openai_compatible.py,sha256=fJFCUz_tyhBG7STItvfNybOq7V620k47_VgSbizRDWU,19441
+review_sensei/providers/openrouter.py,sha256=8gFslaz82oOtDxG58o1-wdn9IH2T_bKjeJE738gphJ8,23224
+review_sensei/providers/profiles.py,sha256=_vhQPNMBT2x-6AU-xVqM0lqTlT2cTiXMv5xwb5-C2EM,9352
+review_sensei/providers/registry.py,sha256=_YifbpFtgsARRF5l1VJsyQgovJgjrgvB_VLT-SUAE9k,12088
+review_sensei/providers/routing.py,sha256=-EttO963m9Gh3E5hlF72EvyFDN9qMabPmm05b3m5QH0,7064
+review_sensei/providers/transport.py,sha256=9QhvHM33gYE7DbmXNxhfBoVDsa7Hx3audaxzRd_7pS4,2937
+review_sensei/release_manifest.py,sha256=a7pPhr0YVBCrzC6Q8JC4HiqTD3aBUyTYV4Y0827U5TA,37775
+review_sensei/schemas.py,sha256=DG0Z2VsZUKRaoLNquemWSsGT8XFavUy1ot-W3ADowWw,2736
+review_sensei/schemas/blocker-admission.schema.json,sha256=M3nBzti7M5yAWKFDcsfdJ3TsPbiJrEEn2Zj4yYUDBPQ,2800
+review_sensei/schemas/canary-binding.schema.json,sha256=065T5GIYG16P9S22HEKrHeRg6unHu8WXedbJxwUYBgk,794
+review_sensei/schemas/candidate-finding.schema.json,sha256=P_FM6pbwdOZGZPAiI_E_ZSmxDfScu2u_2RPzpkdIdmk,1982
+review_sensei/schemas/channel-promotion.schema.json,sha256=LNQgvm6IxntfXrpEnoVz4aL_7_ALo8syiVYEFrfxHYw,1085
+review_sensei/schemas/compatibility-manifest.schema.json,sha256=RIsEo5Os6obelQ078cySd3BHJxGgAySCQg0f0KHRpzY,1858
+review_sensei/schemas/concurrency-plan.schema.json,sha256=YtHPlfShzFPLE4sQMEn7TogW3souwiKC6d4yd-kgFRY,830
+review_sensei/schemas/convergence-sequence-report.schema.json,sha256=FWNUHgdlqlu9SOazQsEVxhBqFY4P_uLynb4OsI8VE1E,2323
+review_sensei/schemas/conversation-reply.schema.json,sha256=bKHxO1CX_g_VrWDb1e0GKPGIUzMPORchHqrT805yPXA,682
+review_sensei/schemas/coverage-manifest.schema.json,sha256=rrbvq1YYNt-c4YPLmdmrzWo_Au-rDu-e35eg3NdZzQQ,2765
+review_sensei/schemas/evaluation-corpus.schema.json,sha256=0W3-zgmVFa5AjDSs385szKNt4DX_D_ZdrE9ur3Nrnz0,7905
+review_sensei/schemas/evaluation-report.schema.json,sha256=kPLXcJv3XeRFNW__VnQzrnovXHaKfS4-ZevfLDgV1xY,10012
+review_sensei/schemas/later-finding-classification.schema.json,sha256=YxsogG9aN4Kbm7__lJpvhOMBbuRe8NbDKS4GnJf4Kl4,2092
+review_sensei/schemas/learning-entry.schema.json,sha256=QtbNYK_eIPmXos4TlYbCFL8EvYAqOx-dxUXDM5wBGQ4,1782
+review_sensei/schemas/learning-feedback.schema.json,sha256=SbNqRSdA82pV5a8Zbzbnjm7U7sj4s9epqRPlJ7-lgAg,1925
+review_sensei/schemas/learning-proposal.schema.json,sha256=FVYG4sFWGil_66-6NlliFCk2LpiCfJn-yTrY4dN-b0E,644
+review_sensei/schemas/observed-convergence-report.schema.json,sha256=amIk4vhjbfn_Zpyz3_qRWHsUJ3M3RAxrXWAvx_K3ipg,5228
+review_sensei/schemas/openrouter-qualification.schema.json,sha256=wIgsZxwvAsKuptmjfsD_8UitPcl-DLAzxjlCD-pZyx8,4203
+review_sensei/schemas/promotion-record.schema.json,sha256=CHJ99AjwioPthygjTimPjXNuG2rriW4DFZw54fa1rfs,2402
+review_sensei/schemas/recovery-artifact.schema.json,sha256=ZP-C_E_nQlFsuObxFn2O0iJLMbDMgNWAF7h4BpPyM3g,1254
+review_sensei/schemas/review-category.schema.json,sha256=Rni93qGN2LyJSrHRF-GcgVdAjqXGnDU5yjIjG2wn7pI,2189
+review_sensei/schemas/review-comment.schema.json,sha256=0A9J_vzYnMUpKsv9wZUGGnCd1ryP-8uksPEbSJWj8Xo,1140
+review_sensei/schemas/review-convergence-policy.schema.json,sha256=E4pGr4CNMzvBUdJ7hTFJy4tzm7iolitG0vClqsj1rcU,1348
+review_sensei/schemas/review-result.schema.json,sha256=jzQMoZIMw2HVC4EhUPfcSuScCd48URPfq6DGX9KnHZM,4899
+review_sensei/schemas/review-round-decision.schema.json,sha256=4Nd306X6_pU4co4papD0nv6D0K5-n3LefGbUYC9M1Ow,1594
+review_sensei/schemas/review-stage.schema.json,sha256=H5CDiC-ngQXJpVomEvd8YPP1tgyPgwo_k3atK4Vjev0,4707
+review_sensei/schemas/review-transaction.schema.json,sha256=KWtsfa0p0CmPx--cB91XpDizF2Z6N2Ehg5YMlRqJwN8,2215
+review_sensei/schemas/run-outcome.schema.json,sha256=pyByHFmQ8yNFPsLD7wccR5lxsWBwzoxNyD77bjNqeU8,1744
+review_sensei/schemas/session-record.schema.json,sha256=DCkPnK8o5va5lkPInHZm6dLyZKnkDKJmNVCD_b06liM,11949
+review_sensei/schemas/stage.schema.json,sha256=WsxbwZFGpRNCyZ6xzWsAQHf8f7YpJLAh9FKRXN_vlao,2900
+review_sensei/schemas/verification-result.schema.json,sha256=Hf765S-aC1Tk4c7KMWaaSf5DMsg6ogmjnHXcjcHE01k,625
+review_sensei/schemas/verification-scope.schema.json,sha256=402Q7OiLHnb41lwb5JkVoKUpCe19fMV9hVhTjsenLFo,2141
+review_sensei/sequence.py,sha256=kUehmk85gtZPSpSsAiUxtK_UVeVZPK6MFwy2av-IGXc,23791
+review_sensei/service.py,sha256=wfU84eab9uXH-ph3JYxcUGWni6oKy1pRLnySqu-thVc,70808
+review_sensei/session.py,sha256=x1VcNk5EF4jjLxPU1D0LVl73E-hDgofmeZEpT6Z8O2o,130802
+review_sensei/stages.py,sha256=rMomi3dvKrEl3yiuy2VbX09_bLu8dWD2rblFH8aJ6yI,24256
+review_sensei/validation.py,sha256=euLaI7cxwTCxcpNRh-F8zGz0BQu8GMG3cA3_1_wtdkU,12390
+review_sensei/verifier.py,sha256=_6cbaoXj8R9hatM7fZav-BgfkSPRyJ4ASmQlZbxF5HA,34913
+review_sensei/workflow.py,sha256=b20eeP9ekbNYGibNCYCcvKSLK7cO3bpCsXiA7y9y33Y,23572