diff --git a/.claude/rules/code-review.md b/.claude/rules/code-review.md
index 11b1107991..73b1a587b0 100644
--- a/.claude/rules/code-review.md
+++ b/.claude/rules/code-review.md
@@ -24,7 +24,7 @@ Project-specific rules layered on top of `CLAUDE.md` for `/quality:agent-review`
 
 ## Critical File Patterns
 
-Files that control auth, routing, Apollo setup, build config, or the CI/review pipeline. Each contributes +3 to risk score (on top of the universal defaults).
+Highest-risk tier — files that control auth, routing, Apollo setup, build config, or the CI/review pipeline. Additive to the universal defaults. _(This file only sorts files into tiers; how each tier maps onto the risk axes is defined by the skill.)_
 
 - `pages/api/auth/[...nextauth].page.ts` — NextAuth handler (OAuth callbacks, session)
 - `pages/api/auth/apiOauthSignIn.ts` — API OAuth sign-in flow
@@ -34,7 +34,7 @@ Files that control auth, routing, Apollo setup, build config, or the CI/review p
 - `pages/_app.page.tsx` — app-level providers and global wiring
 - `next.config.{js,ts}` — build-time config, headers, rewrites, CSP
 - `codegen.ts`, `codegen.*.ts` — GraphQL codegen configuration
-- `package.json` — dependency changes (new packages trigger `## Special Pattern Detection` below)
+- `package.json` — dependency changes (supply-chain surface; the skill's universal signals also flag new/updated dependencies)
 - `.github/workflows/**` — CI/CD workflows
 - `src/lib/apollo/client.ts` — Apollo Client instantiation (auth headers, default options, link chain composition)
 - `src/lib/apollo/link.ts` — Apollo Link chain (dual-server routing, auth token attachment, error handling)
@@ -44,7 +44,7 @@ Files that control auth, routing, Apollo setup, build config, or the CI/review p
 
 ## High-Risk File Patterns
 
-Each contributes +2 to risk score.
+Second tier — sensitive or wide-reach files below Critical.
 
 - `pages/api/Schema/**/*.{ts,graphql}` — REST-proxy resolvers, schemas, and data handlers (silent data-reshaping risk)
 - `pages/api/Schema/**/datahandler*.ts` — response transformation feeding the REST proxy
@@ -58,7 +58,7 @@ Each contributes +2 to risk score.
 
 ## Medium-Risk File Patterns
 
-Each contributes +1 to risk score.
+Third tier — everyday feature, hook, and utility files.
 
 - `src/components/**/*.{ts,tsx}` — feature components (excluding Shared, already High-Risk)
 - `src/hooks/**/*.ts` — custom hooks
@@ -69,7 +69,7 @@ Each contributes +1 to risk score.
 
 ## Low-Risk File Patterns
 
-Zero points. These override or augment the universal Low-Risk defaults.
+Lowest tier — files that may legitimately carry no risk. These override or augment the universal Low-Risk defaults.
 
 - `**/*.test.{ts,tsx}` — test files (content changes only; new test infrastructure should still be reviewed)
 - `public/locales/**/*.json` — translation content
@@ -77,20 +77,6 @@ Zero points. These override or augment the universal Low-Risk defaults.
 - `public/fonts/**`, `public/images/**` — binary assets
 - `**/*.snap` — Jest snapshot files
 
-## Special Pattern Detection
-
-Additional risk modifiers, added to the universal defaults.
-
-- **New package added to `package.json` dependencies/devDependencies:** +2 (supply-chain risk, bundle size impact)
-- **Updated critical package** (`next`, `react`, `@apollo/client`, `@mui/material`, `formik`, `next-auth`, `typescript`, `graphql-codegen`): +3
-- **`yarn.lock` changed without matching `package.json` change:** +1 (likely a resolution drift or lockfile hand-edit)
-- **`.graphql` file changed without verifying `yarn gql` runs cleanly:** +2 (codegen out of sync — runtime errors likely). Note: `.generated.ts` files are not committed; CI regenerates them. The check is that codegen succeeds, not that generated files appear in the PR
-- **New `.graphql` file under `pages/api/Schema/` without matching resolver/dataHandler updates:** +1
-- **New file in `src/hooks/` that uses Apollo hooks without an accompanying test file:** +1 (hooks drive component behavior; untested hooks are landmines)
-- **New file in `src/components/` without an accompanying `*.test.tsx`:** +1
-- **Changes to `next.config.{js,ts}` rewrites, headers, CSP, or image domains:** +2 (affects all pages and external requests)
-- **Changes to `src/lib/apollo/cache.ts` `typePolicies` or `merge` functions:** +2 (can silently corrupt cached data across the app)
-
 ## Agent Triggers
 
 Repo-specific triggers that supplement the skill's minimal universal triggers.
@@ -280,6 +266,8 @@ This is where domain-specific data invariants belong.
 - **Pagination over `nodes`** — aggregating client-side over a single page of `nodes` silently ignores unpaginated data. Prefer server-provided totals
 - **Filter/search state** — when filters change, paginated data must be re-fetched from the first page, not appended to existing pages
 - **Optimistic updates on lists** — adding an item optimistically must place it in the correct sort position, not just at the end
+- **Codegen sync** — when a `.graphql` operation or schema changes, `yarn gql` must run cleanly. `.generated.ts` files are not committed (CI regenerates them), so verify codegen _succeeds_ rather than expecting generated files in the PR; drift causes runtime type errors
+- **New REST-proxy operations** — a new `.graphql` file under `pages/api/Schema/` needs a matching `resolvers.ts` / `datahandler.ts`, or it will fail at runtime
 
 ## Testing Focus Areas
 
diff --git a/.github/workflows/ai-review-auto-approve.yml b/.github/workflows/ai-review-auto-approve.yml
index ca6178a77e..467de4bf47 100644
--- a/.github/workflows/ai-review-auto-approve.yml
+++ b/.github/workflows/ai-review-auto-approve.yml
@@ -77,17 +77,19 @@ jobs:
             const risk_level = String(meta.risk_level || '').toUpperCase().trim();
             const verdict = String(meta.verdict || '').toUpperCase().trim();
 
-            if (isNaN(Number(meta.risk_score))) {
-              console.log(`Warning: risk_score "${meta.risk_score}" is not a number, defaulting to 0`);
-            }
             if (isNaN(Number(meta.blockers))) {
               console.log(`Warning: blockers "${meta.blockers}" is not a number, defaulting to 0`);
             }
 
-            const risk_score = Number(meta.risk_score) || 0;
             const blockers = Number(meta.blockers) || 0;
             const dismissed = Number(meta.dismissed) || 0;
-            console.log(`Agent review found — Risk: ${risk_level} (${risk_score}/10), Blockers: ${blockers}, Verdict: ${verdict}, Dismissed: ${dismissed}`);
+
+            // Axis risk model. These are for display only.
+            const axes = meta.axes && typeof meta.axes === 'object' ? meta.axes : null;
+            const danger = axes && !isNaN(Number(axes.danger)) ? Number(axes.danger) : null;
+            const effort = axes && !isNaN(Number(axes.effort)) ? Number(axes.effort) : null;
+            const riskDetail = `danger ${danger ?? '?'}/6, effort ${effort ?? '?'}/6`
+            console.log(`Agent review found — Risk: ${risk_level} (${riskDetail}), Blockers: ${blockers}, Verdict: ${verdict}, Dismissed: ${dismissed}`);
 
             // Gate 1: Only LOW or MEDIUM risk
             if (!['LOW', 'MEDIUM'].includes(risk_level)) {
@@ -133,7 +135,7 @@ jobs:
               const bodyLines = [
                 '## AI Review Auto-Approval',
                 '',
-                `**Risk Level:** ${risk_level} (${risk_score}/10)`,
+                `**Risk Level:** ${risk_level} (${riskDetail})`,
                 `**Verdict:** ${verdictDesc}`,
                 '',
                 'This PR was auto-approved because:',