diff --git a/docs/skills.md b/docs/skills.md
index afbac0d2de..107de805d7 100644
--- a/docs/skills.md
+++ b/docs/skills.md
@@ -449,6 +449,14 @@ That is the point of `/review`.
 I do not want flattery here.
 I want the model imagining the production incident before it happens.
 
+### Repo-owned review context
+
+`/review` already uses `.claude/skills/review/checklist.md` for the actual rubric.
+
+If you need to calibrate `/review` for your specific repo — scope rules, high-risk paths, trust boundaries, auto-fix boundaries, escalation rules, or external consumers — add a `## Review` section to your project's `CLAUDE.md`.
+
+`/review` reads it before scope drift detection and uses it for risk and scope calibration. Skips silently if the section doesn't exist.
+
 ---
 
 ## `/investigate`
diff --git a/examples/CLAUDE.md.review.example b/examples/CLAUDE.md.review.example
new file mode 100644
index 0000000000..57fa897ff9
--- /dev/null
+++ b/examples/CLAUDE.md.review.example
@@ -0,0 +1,15 @@
+# Example: `## Review` section for CLAUDE.md
+
+Add this section to your project's `CLAUDE.md` to give `/review` repo-specific context.
+
+```markdown
+## Review
+
+- Source of truth for intent: if the PR or commit message links a GitHub issue / Linear ticket, read it before deciding scope drift.
+- High-risk paths: `db/migrate/**`, `app/services/billing/**`, `config/initializers/auth*`.
+- Never AUTO-FIX without explicit approval: money movement, auth/session logic, customer-visible API contract changes.
+- External consumers exist outside this repo: mobile app, webhooks, warehouse sync jobs.
+- Known hotspot: `app/services/reconciliation/**` has intentionally defensive duplication; do not flag it as dead code without tracing the call sites.
+- If a diff touches a high-risk path and the stated intent is silent about it, call out probable scope drift explicitly.
+- If auth or billing code changes, prefer ASK over AUTO-FIX unless the fix is obviously mechanical and reversible.
+```
diff --git a/review/SKILL.md b/review/SKILL.md
index 912e1f3ec5..f551ffd07c 100644
--- a/review/SKILL.md
+++ b/review/SKILL.md
@@ -288,6 +288,16 @@ You are running the `/review` workflow. Analyze the current branch's diff agains
 
 ---
 
+## Step 1.25: Read project review context (optional)
+
+Before the main review, check `CLAUDE.md` for a `## Review` section. If present, read it as additive repo-specific context — not a replacement for `.claude/skills/review/checklist.md`.
+
+Use it for scope calibration (source of truth for intent, ticketing conventions), risk calibration (high-risk paths, trust boundaries), escalation rules (who to involve for auth/billing changes), and auto-fix boundaries (areas that should never be AUTO-FIXed without explicit approval).
+
+If this context names an accessible ticketing source of truth, use it during Scope Drift Detection in Step 1.5. If no `## Review` section exists, skip silently.
+
+---
+
 ## Step 1.5: Scope Drift Detection
 
 Before reviewing code quality, check: **did they build what was requested — nothing more, nothing less?**
diff --git a/review/SKILL.md.tmpl b/review/SKILL.md.tmpl
index 712b91a905..645ec68cc9 100644
--- a/review/SKILL.md.tmpl
+++ b/review/SKILL.md.tmpl
@@ -37,6 +37,16 @@ You are running the `/review` workflow. Analyze the current branch's diff agains
 
 ---
 
+## Step 1.25: Read project review context (optional)
+
+Before the main review, check `CLAUDE.md` for a `## Review` section. If present, read it as additive repo-specific context — not a replacement for `.claude/skills/review/checklist.md`.
+
+Use it for scope calibration (source of truth for intent, ticketing conventions), risk calibration (high-risk paths, trust boundaries), escalation rules (who to involve for auth/billing changes), and auto-fix boundaries (areas that should never be AUTO-FIXed without explicit approval).
+
+If this context names an accessible ticketing source of truth, use it during Scope Drift Detection in Step 1.5. If no `## Review` section exists, skip silently.
+
+---
+
 ## Step 1.5: Scope Drift Detection
 
 Before reviewing code quality, check: **did they build what was requested — nothing more, nothing less?**