diff --git a/CHANGELOG.md b/CHANGELOG.md
index a06257f..e9e40f9 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -2,6 +2,8 @@
 
 ## Unreleased
 
+- **Skill relay-depth guidance matches policy enforcement (T696).** The skill now says ask allows 6 hops and collaborate, autonomous and yolo have no limit. A regression test checks the documented limits against `LEVEL_MAX_HOP`.
+
 ## 0.5.47 (2026-10-16)
 
 - **Service-hosted OpenCode sessions stay wakeable after MCP eviction (T688).** OpenCode stops an idle session's MCP about hourly, which releases its lease. T659 kept standalone sessions wakeable through their plugin's poll; a session hosted by the shared `opencode serve --service` is now a wake target too while that service process (same pid and birth) lives, and the wake posts to the service with `resume: true` and a verified receipt. A lease held by another session of the agent still blocks it. docs/OPENCODE.md recommends plain `opencode` (one service per machine, every project) for long-lived agents.
diff --git a/skill/SKILL.md b/skill/SKILL.md
index 635f824..ffdc4ae 100644
--- a/skill/SKILL.md
+++ b/skill/SKILL.md
@@ -250,7 +250,7 @@ wake only queues a short pointer to the inbox; it never means the recipient read
 - Check the receipt's `recipients[].state` (`live-wake` or `live-next-prompt`) rather than assuming the send woke anyone.
 
 Relay depth counts only what you read from agents other than the recipients: answering the agent you heard from never
-adds depth, forwarding to someone else does (ask allows 6, collaborate 20, autonomous/yolo no limit; the policy line shows it).
+adds depth, forwarding to someone else does (ask 6, collaborate no limit, autonomous no limit, yolo no limit; the policy line shows it).
 
 ## Autonomy probes (`[mbx-probe]`)
 
diff --git a/test/skill-relay-depth.test.ts b/test/skill-relay-depth.test.ts
new file mode 100644
index 0000000..e5a3818
--- /dev/null
+++ b/test/skill-relay-depth.test.ts
@@ -0,0 +1,11 @@
+import { test } from "node:test";
+import assert from "node:assert/strict";
+import { readFileSync } from "node:fs";
+import { LEVEL_MAX_HOP } from "../src/policy.ts";
+
+test("the skill relay-depth limits agree with policy enforcement", () => {
+  const skill = readFileSync(new URL("../skill/SKILL.md", import.meta.url), "utf8");
+  const limits = Object.entries(LEVEL_MAX_HOP)
+    .map(([level, limit]) => `${level} ${Number.isFinite(limit) ? limit : "no limit"}`).join(", ");
+  assert.ok(skill.includes(`(${limits}; the policy line shows it).`), `skill must document: ${limits}`);
+});