Release the adaptive review controller with operating guidance

.agents/plugins/marketplace.json

@@ -0,0 +1,10 @@
+{
+  "name": "code-review-c",
+  "interface": {"displayName": "code-review-c"},
+  "plugins": [{
+    "name": "code-review-c",
+    "source": {"source": "local", "path": "./"},
+    "policy": {"installation": "AVAILABLE", "authentication": "ON_INSTALL"},
+    "category": "Developer Tools"
+  }]
+}

.codex-plugin/plugin.json

@@ -0,0 +1,9 @@
+{
+  "name": "code-review-c",
+  "version": "0.1.0",
+  "description": "Adaptive focused Codex reviews with measured model onboarding.",
+  "author": {"name": "Dmitry Kropachev"},
+  "license": "MIT",
+  "repository": "https://github.com/dk-code-factory/code-review-c",
+  "skills": "./codex/skills/"
+}

.github/workflows/ci.yml

@@ -0,0 +1,33 @@
+name: CI
+
+on:
+  push:
+  pull_request:
+
+permissions:
+  contents: read
+
+concurrency:
+  group: ci-${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  offline-checks:
+    name: Python ${{ matrix.python }}
+    runs-on: ubuntu-latest
+    timeout-minutes: 10
+    strategy:
+      fail-fast: false
+      matrix:
+        python: ['3.11', '3.12', '3.13', '3.14', '3.15']
+    steps:
+      - uses: actions/checkout@d23441a48e516b6c34aea4fa41551a30e30af803 # v6
+        with:
+          persist-credentials: false
+      - uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6
+        with:
+          python-version: ${{ matrix.python }}
+      - name: Check package
+        run: python scripts/check.py
+      - name: Run offline tests
+        run: python scripts/test.py

NOTICE

@@ -0,0 +1,13 @@
+code-review-c
+Copyright (c) 2026 Dmitry Kropachev
+Licensed under the MIT License; see LICENSE.
+
+The generic evidence, inspection, and aspect guidance is adapted from:
+https://github.com/dk-code-factory/code-review-ad
+commit 5900df7e96463f0fc68a5c3c0d5e0dcfdf36cd95
+Copyright (c) 2026 Dmitry Kropachev
+MIT License (the complete notice and permission text is retained in LICENSE).
+
+The adaptive controller, standard-library helpers, backend briefs, onboarding
+harness and tests are new code-review-c work. No installed external skills,
+private rules, model profiles, benchmark results or credentials are included.

PROVENANCE.json

@@ -0,0 +1,44 @@
+{
+  "version": "0.1.0",
+  "sources": [
+    {
+      "repository": "https://github.com/dk-code-factory/code-review-ad",
+      "commit": "5900df7e96463f0fc68a5c3c0d5e0dcfdf36cd95",
+      "license": "MIT",
+      "notice": "NOTICE",
+      "sourceHashes": {
+        "common/inspection.md": "80582feb0b2ec1db49eb0618b959fc7395c4a451c6420a205cf7450e3b125cce",
+        "common/reporting.md": "89edeba76749bc006420d6d7f33440ee9c8df71432836c72c2e0e76c698cd5e6",
+        "common/codex/passes/toolkit.md": "5033c2cf052084219a0e7ff50770d9fdecd2a75fcdbf0a92357f154469a85c69",
+        "common/codex/passes/standards.md": "687efa713215b1acb3f838252873309d504e21abb6eee5491d70015e0f205393",
+        "common/codex/passes/spec.md": "6eeb4841110941a677012d546e629131e8182f7ce600fc5d1ba9438e2742e349",
+        "common/codex/passes/performance.md": "4a776a01a7d8742fe980ee36c2156797c68863e4ec386785a85c1ed8d4663cdc"
+      },
+      "adaptedInto": [
+        "codex/skills/code-review-c/references/evidence.md",
+        "codex/skills/code-review-c/references/generic.md",
+        "codex/skills/code-review-c/references/aspects/"
+      ]
+    }
+  ],
+  "ownedChanges": {
+    "description": "New adaptive controller, focused backend guidance, deterministic inventory/planning/gates, onboarding harness and offline tests",
+    "paths": [
+      "codex/skills/code-review-c",
+      "scripts",
+      "tests"
+    ]
+  },
+  "documentation": [
+    {
+      "url": "https://developers.openai.com/blog/eval-skills",
+      "accessed": "2026-10-24",
+      "use": "Behavioral evaluation method; no copied code"
+    },
+    {
+      "url": "https://developers.openai.com/codex/noninteractive",
+      "accessed": "2026-10-24",
+      "use": "Isolated exec flags and structured output"
+    }
+  ]
+}

README.md

@@ -1,8 +1,70 @@
 # code-review-c
 
-Standalone adaptive Codex review skill. MIT licensed. The implementation lives on
-`feat/adaptive-review-controller`; `main` contains the public scaffold.
+Standalone adaptive Codex review skill. Version 0.1.0, MIT licensed.
 
 Review quality comes first. Initial workload limits are provisional, not measured
 model capacities. The first release supplies onboarding tools and runs no paid
 calibration campaigns.
+
+## Use
+
+Install the stable `v0.1.0` tag into your Codex skills directory:
+
+```sh
+git clone --branch v0.1.0 --depth 1 https://github.com/dk-code-factory/code-review-c.git
+skill_parent="${CODEX_HOME:-$HOME/.codex}/skills"
+mkdir -p "$skill_parent"
+test ! -e "$skill_parent/code-review-c" && cp -R code-review-c/codex/skills/code-review-c "$skill_parent/"
+```
+
+If that skill already exists, move the old directory aside before installing;
+keep it for rollback. Alternatively, add the tagged checkout as a local Codex
+plugin marketplace. The package contains exactly one skill, `code-review-c`.
+It needs no sibling checkout, external review plugin, or build step. Select it
+in Codex and request:
+
+- “Review PR URL with code-review-c.”
+- “Preview a strategy for base REV and head REV.”
+- “Prepare model onboarding; do not run trials.”
+
+Linux is the first-release support target. CI covers CPython 3.11–3.15; other
+platforms and Python versions are unverified. Git and native Codex agents support
+review/preview; authenticated `gh` supports GitHub PR pinning. Docker is needed
+only for isolated onboarding trials and container integration checks. Models and effort inherit the host selection. A missing or
+incompatible profile uses a provisional 58,000-byte total visible-input budget.
+No measured profiles are distributed. Binary, submodule and type changes remain
+explicit gaps requiring separate inspection; indivisible oversized hunks also remain
+gaps. Dirty worktrees are recorded, but evidence comes only from pinned commits.
+
+The controller identifies semantic relationships and applicable aspects; deterministic
+helpers enforce exact hunk/aspect assignment, byte budgets, task dependencies and
+whole-PR essential gates. Tasks queue within native concurrency. Security and serious
+resource/performance checks remain essential after correctness findings. Optional
+improvements open only after complete, verified, clean essential coverage.
+
+Read the [skill](codex/skills/code-review-c/SKILL.md),
+[controller procedure](codex/skills/code-review-c/references/controller.md),
+[record contracts](codex/skills/code-review-c/references/contracts.md), and
+[onboarding procedure](codex/skills/code-review-c/references/onboarding.md), and
+[operating guide](codex/skills/code-review-c/references/operations.md).
+
+## Development
+
+```sh
+python3 scripts/check.py
+python3 scripts/test.py
+```
+
+Runtime and tests use the Python standard library. Tests simulate reviewers and
+calibration receipts; they make no model calls. Onboarding preparation freezes
+fixtures and candidate strategies without inference. Model trials require explicit
+call/time limits and an immutable Docker image with Codex installed. Trial containers
+see only pinned source/evidence and temporary Codex authentication, keeping answer
+keys outside the reviewer filesystem. Three matched repetitions, held-out fixtures,
+critical-defect recall, precision and completion gates precede profile promotion.
+
+Keep review artifacts, campaign data, local profiles, credentials, private rules,
+caches and benchmark results outside the package and reviewed repository. Provenance
+for adapted review guidance is pinned in [PROVENANCE.json](PROVENANCE.json), with
+upstream license notices retained in [NOTICE](NOTICE). This package does not change
+the source `code-review-ad` repository or its configured reviewers.

codex/skills/code-review-c/SKILL.md

@@ -0,0 +1,40 @@
+---
+name: code-review-c
+description: Review a pull request or pinned local revisions with adaptive focused Codex agents, preview a review strategy, or onboard a model using isolated evaluation campaigns.
+---
+
+Use this session as the controller. Resolve bundled paths from this skill directory,
+not the reviewed repository. Python 3.11+, Git, and native Codex agents are required;
+PR pinning also uses authenticated `gh`. No other skill or plugin is required.
+
+Choose the requested mode:
+
+- Review or preview: read [the controller procedure](references/controller.md).
+  Preview stops after validated planning and reports the proposed strategy and gaps.
+- Onboard a model or import prior measurements: read [the onboarding procedure](references/onboarding.md).
+  Preparation and comparison run no models. Paid trials require explicit limits.
+
+In review mode, use fresh native agents without inherited conversation for each
+task. Reviewers do not spawn agents. Inherit the host-selected model and reasoning
+effort; never set reviewer overrides. Resolve the actual reviewer identity from host
+metadata when available. The controller's identity does not establish it. Unknown
+identity, changed instructions, or missing compatible measurements require the
+provisional strategy: at most **58,000 bytes total visible input per task**, including
+patch, instructions, evidence, framing, and host-visible instructions. This is an
+initial safety budget, not an established capacity or a token count.
+
+Cover every changed hunk and all essential aspects before optional improvements.
+Keep security and serious resource/performance checks active when other reviewers
+find defects. Verify essential candidates before opening later stages. A verified
+essential defect, unresolved essential finding, or incomplete essential coverage
+blocks optional reviews across the entire PR. Queue remaining work within available
+host concurrency; never reduce coverage to fit the number of agents.
+
+Use measured profiles only for their exact identity, instruction hash, language,
+domain, and primary review aspect. Prefer quality; use tokens and latency to break
+quality ties. Keep all campaign data, traces, local profiles, and review artifacts
+outside this plugin and the reviewed repository.
+
+Return verified defects in priority order, then any separately labeled optional
+improvements, material coverage gaps, and the reason optional stages were skipped.
+Keep reports concise. Review does not authorize code edits or GitHub publication.

codex/skills/code-review-c/agents/openai.yaml

@@ -0,0 +1,4 @@
+interface:
+  display_name: "Adaptive Code Review"
+  short_description: "Focused reviews with measured model profiles"
+  default_prompt: "Use $code-review-c to review this branch against its base."

codex/skills/code-review-c/references/aspects/concurrency.md

@@ -0,0 +1,7 @@
+# Concurrency
+
+Trace writers/readers, synchronization, ownership transfer, queue progress, and shutdown.
+For every new wait, enumerate what settles it and whether peer I/O, cancellation or
+abort is sequenced after the wait. Compare ordering with base. Verify lifecycle guards
+cover setup and alternate writers; supported compile/runtime conditions matter. Show
+a reachable interleaving and consequence rather than a hypothetical race.

codex/skills/code-review-c/references/aspects/contracts.md

@@ -0,0 +1,7 @@
+# Interface contract
+
+Inspect only the demonstrated relationship named in this task. Compare caller and
+callee/schema/transport representations at the pinned revisions: types, names,
+nullability, ordering, errors, ownership, authorization and completion semantics.
+Use both endpoint hunk IDs and supplied evidence. Establish a concrete mismatch or
+state that inspection was incomplete; shared directories are not evidence of a contract.

codex/skills/code-review-c/references/aspects/correctness.md

@@ -0,0 +1,6 @@
+# Correctness
+
+Trace changed control flow, state transitions, boundary values, and supported failure
+paths against base. Check early returns that bypass updates, ordering of completion,
+stale state after failure, and assumptions made by callers. Read changed comments for
+factual contradictions with those paths. Report supported defects, not alternative designs.

codex/skills/code-review-c/references/aspects/data-consistency.md

@@ -0,0 +1,6 @@
+# Data consistency
+
+Follow transaction and state boundaries across persistence, caches, queues and APIs.
+Check atomicity, isolation assumptions, retry/idempotency keys, migration ordering,
+partial failure, and stale state while failures continue. Identify the concrete
+invariant, reachable sequence, and observable violation, including recovery behavior.

codex/skills/code-review-c/references/aspects/requirements.md

@@ -0,0 +1,7 @@
+# Explicit requirements
+
+Compare changed behavior with the pinned PR, supplied issues, user requirements and
+acceptance criteria. Preserve quantifiers, configurations, and timing promises. Map
+claims to actual branches and test stimuli. If no meaningful spec exists, state
+`No spec available` as a gap; do not invent requirements. Distinguish a newly broken
+promise from unchanged pre-existing behavior.

codex/skills/code-review-c/references/aspects/resources.md

@@ -0,0 +1,7 @@
+# Serious resource and performance defects (essential)
+
+Trace acquisition/release, bounded queues and allocations, cancellation, error cleanup,
+retry loops, shutdown, and work per request/item/retry. Show a realistic trigger and
+consequential leak, exhaustion, unbounded growth, hot-path regression, or liveness
+failure compared with base. Quantify work from code where possible. This check stays
+active after other findings; optional scalability advice cannot replace it.

codex/skills/code-review-c/references/aspects/rules.md

@@ -0,0 +1,7 @@
+# Mandatory repository rules
+
+Apply only documented guidance with matching scope. Cite the exact rule and predicate
+violated, including exceptions. Compare changed setup/cleanup and public interfaces
+with specific established conventions when the repository makes them mandatory.
+Report evidenced maintenance defects only when named sites, coupled state, API
+obligations or supported failure modes demonstrate a coordination consequence.

codex/skills/code-review-c/references/aspects/scalability.md

@@ -0,0 +1,7 @@
+# Scalability improvements (optional)
+
+After the whole-PR essential gate is clean, inspect the named changed behavior for
+growth in work, I/O, memory or contention under a stated realistic workload. Explain
+the benefit and assumptions of an improvement. Do not present speculative scale as
+an existing defect. If a serious defect appears, return incomplete and ask the
+controller to reopen essential coverage.

codex/skills/code-review-c/references/aspects/security.md

@@ -0,0 +1,7 @@
+# Security
+
+Follow changed trust boundaries: identity, authorization, tenant/object ownership,
+untrusted input, filesystem/process/query sinks, credentials and logs. Verify that
+validation occurs on every reachable supported path, including retries and alternate
+APIs. Establish attacker-controlled input, a reachable sink, and impact. Keep this
+check active even when correctness reviewers have findings; omit generic hardening.

codex/skills/code-review-c/references/aspects/simplification.md

@@ -0,0 +1,7 @@
+# Simplification improvements (optional)
+
+After the whole-PR essential gate is clean, examine only the changed behavior named
+by the controller. Identify concrete duplicated work, unnecessary state or coordination
+and the supported behavior a simpler design must preserve. Avoid style preferences,
+rewrites without benefit, or invented future requirements. New suspected defects
+reopen essential coverage rather than becoming suggestions.

codex/skills/code-review-c/references/aspects/tests.md

@@ -0,0 +1,8 @@
+# Test defects
+
+Map changed behavior to actual test stimuli. Check assertions do not already hold
+before the action, both sides of interfaces are exercised, failure paths are reached,
+waits have deadlines, and cleanup survives failed assertions/exceptions. Treat missing
+deadlines and skipped cleanup as distinct mechanisms. Inspect build/configuration
+coverage claimed by changed tests and comments. Missing tests alone is not a defect
+without a claimed or mandatory coverage obligation or a demonstrable wrong assertion.

codex/skills/code-review-c/references/backend/c.md

@@ -0,0 +1,6 @@
+# C backend
+
+Trace bounds, integer overflow, ownership, allocation failure, aliasing, lifetime,
+and all cleanup exits. Compare ABI/layout and compile-time feature guards with
+callers. For concurrent paths verify synchronization and shutdown progress under
+the actual memory model and supported build configurations.

codex/skills/code-review-c/references/backend/common.md

@@ -0,0 +1,10 @@
+# Backend evidence lenses
+
+Use only lenses relevant to the task's primary purpose. Trace API input/output and
+compatibility; authentication versus object/tenant authorization; persistence and
+transaction boundaries; schema/migration compatibility during mixed-version deploys;
+async ownership, retries, cancellation and shutdown; and resource cleanup after every
+exit. For retries inspect stored state used by the next attempt, idempotency, backoff
+and whether progress needs external change. For pooled resources inspect expiry and
+reconfiguration. Evidence of an actual contract violation is required; retained desired
+state or a delayed retry can be intentional.

codex/skills/code-review-c/references/backend/cpp.md

@@ -0,0 +1,6 @@
+# C++ backend
+
+Trace RAII/lifetime and reference/iterator invalidation, move/copy ownership, exception
+safety, bounds and integer conversion. Check new overrides against declarations and
+supported compile guards. Follow futures, coroutine cancellation, pooled resources,
+retry state and shutdown ordering; demonstrate progress or a reachable blocked path.

codex/skills/code-review-c/references/backend/csharp.md

@@ -0,0 +1,6 @@
+# C# backend
+
+Trace Task ownership, async void boundaries, CancellationToken propagation, blocking
+on async work, and disposal/IAsyncDisposable. Inspect DI scope and DbContext lifetime,
+transaction boundaries, nullable contracts and serializer behavior. Check authorization
+on alternate endpoints and background work as well as the main request path.

codex/skills/code-review-c/references/backend/go.md

@@ -0,0 +1,6 @@
+# Go backend
+
+Trace goroutine exits, context cancellation, channel ownership, close/send order,
+and WaitGroup progress. Check deferred cleanup placement, HTTP bodies, rows/errors,
+transaction rollback, and error wrapping at callers. Compare shared state access and
+slice/map aliasing under the module's actual Go version.

codex/skills/code-review-c/references/backend/java.md

@@ -0,0 +1,6 @@
+# Java backend
+
+Trace executor/future ownership, interruption, blocking under locks, and exception
+translation. Check try-with-resources and transaction/session lifetimes; equality,
+nullability, collection mutation, and publication under actual concurrency. Verify
+framework proxy/annotation boundaries when they control transactions or authorization.

codex/skills/code-review-c/references/backend/javascript.md

@@ -0,0 +1,6 @@
+# JavaScript backend
+
+Trace promise ownership, rejected async callbacks, cancellation/AbortSignal propagation,
+event listeners/timers and stream backpressure. Check coercion, serialization and numeric
+limits at APIs and storage; authorization must precede side effects on every route.
+Distinguish a returned promise from fire-and-forget work using actual callers.

codex/skills/code-review-c/references/backend/php.md

@@ -0,0 +1,6 @@
+# PHP backend
+
+Check request/session and long-lived worker state separately; loose comparisons,
+array/key coercion and null behavior at boundaries; prepared query parameters and
+authorization before side effects. Trace transaction rollback, exception handling,
+stream cleanup and retries under the pinned PHP/framework version.

codex/skills/code-review-c/references/backend/python.md

@@ -0,0 +1,6 @@
+# Python backend
+
+Check async tasks are awaited/owned and cancellation reaches cleanup; context-manager
+and generator lifetimes; shared mutable defaults/state; transaction/autocommit behavior;
+and sync I/O inside event-loop paths. Verify exceptions do not silently turn failed
+operations into success. Use the actual framework/runtime version from pinned evidence.

codex/skills/code-review-c/references/backend/rust.md

@@ -0,0 +1,6 @@
+# Rust backend
+
+Trace future polling/ownership, dropped tasks, cancellation safety, lock guards across
+awaits, channels and shutdown. Check error conversions and their display text against
+all producing paths. Inspect unsafe contracts, lifetimes, atomics, transaction guards,
+and pooled resource expiry/retry state under supported features and targets.