diff --git a/CHANGELOG.md b/CHANGELOG.md
index 095253f..3a3bbc0 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -4,7 +4,10 @@ Notable changes to `@questpie/agent-board` (CLI binaries: `agent-board`, `agent`
 
 ## Unreleased
 
+## 0.7.0 — 2026-06-14
+
 - **Design QA — token scan**: `agent-board-design-qa` gained a second embedded primitive that flags off-token values. It learns the repo's design tokens at runtime by probing every `:root` CSS variable through a throwaway element (works on any CSS-variable design system, no per-stack config) and flags off-palette colors, off-token box-shadows, and off-token border-radii. Spacing is intentionally checked only against a real scale, never a generic 4px heuristic.
+- **Design QA — prefer `qprobe design`**: when `@questpie/probe` is installed, the skill now prefers its `qprobe design <url>` command (the same scanners run as tested code) and falls back to the embedded scans otherwise.
 
 ## 0.6.0 — 2026-06-14
 
diff --git a/docs/internal/design-qa-skill.md b/docs/internal/design-qa-skill.md
index 8dd3315..f0d8390 100644
--- a/docs/internal/design-qa-skill.md
+++ b/docs/internal/design-qa-skill.md
@@ -49,6 +49,8 @@ This skill needs three capabilities from whatever harness you are on — Claude
 
 If none is available, wiring one up is the first step; the rest of the loop assumes these three.
 
+If `@questpie/probe` (qprobe) is installed, skip manual injection: `qprobe design <url> --viewport 375,768,1280` launches headless Chromium and runs the geometry and token scans below as a tested command, printing the same JSON findings. Prefer it when available; the embedded scans are the portable fallback for harnesses without qprobe.
+
 ## Loop
 
 1. Target. Read the task and its spec: `agent-board show <task-id>`, then `agent-board spec cat <spec-id>`. Note the required screens, states, and breakpoints. If a visual reference exists (a mockup via `agent-board web`, a Figma export, or a screenshot), keep its image for step 7 — the reference is just an image; where it comes from is your call.
diff --git a/package.json b/package.json
index aa14cd8..0c80d79 100644
--- a/package.json
+++ b/package.json
@@ -1,6 +1,6 @@
 {
 	"name": "@questpie/agent-board",
-	"version": "0.6.0",
+	"version": "0.7.0",
 	"description": "Markdown task board and execution contract for coding agents",
 	"license": "MIT",
 	"repository": {
diff --git a/src/skills.ts b/src/skills.ts
index b176c60..82f9c87 100644
--- a/src/skills.ts
+++ b/src/skills.ts
@@ -902,6 +902,8 @@ This skill needs three capabilities from whatever harness you are on — Claude
 
 If none is available, wiring one up is the first step; the rest of the loop assumes these three.
 
+If \`@questpie/probe\` (qprobe) is installed, skip manual injection: \`qprobe design <url> --viewport 375,768,1280\` launches headless Chromium and runs the geometry and token scans below as a tested command, printing the same JSON findings. Prefer it when available; the embedded scans are the portable fallback for harnesses without qprobe.
+
 ## Loop
 
 1. Target. Read the task and its spec: \`agent-board show <task-id>\`, then \`agent-board spec cat <spec-id>\`. Note the required screens, states, and breakpoints. If a visual reference exists (a mockup via \`agent-board web\`, a Figma export, or a screenshot), keep its image for step 7 — the reference is just an image; where it comes from is your call.