diff --git a/.claude/settings.json b/.claude/settings.json
new file mode 100644
index 000000000..569e840a8
--- /dev/null
+++ b/.claude/settings.json
@@ -0,0 +1,11 @@
+{
+  "extraKnownMarketplaces": {
+    "dock": {
+      "source": { "source": "github", "repo": "docknetwork/dock-claude-skills" },
+      "autoUpdate": true
+    }
+  },
+  "enabledPlugins": {
+    "dock-skills@dock": true
+  }
+}
diff --git a/.claude/skills/example-skill/SKILL.md b/.claude/skills/example-skill/SKILL.md
new file mode 100644
index 000000000..d908d447f
--- /dev/null
+++ b/.claude/skills/example-skill/SKILL.md
@@ -0,0 +1,20 @@
+---
+name: example-skill
+description: "TEMPLATE — copy this directory, rename it, and write a real trigger here before use. Left disabled so it never fires as-is."
+disable-model-invocation: true
+---
+
+# example-skill
+
+Boilerplate for a repo-specific Claude Code skill. Copy this directory to
+`.claude/skills/<your-skill-name>/`, then:
+
+1. Rename `name` above to match the directory.
+2. Replace `description` with the real trigger: the situations that should make
+   an agent reach for this skill, phrased as explicit branches ("use when the
+   user says X" / "use after Y fails"), not a vague "use for X".
+3. Remove `disable-model-invocation: true` once the skill should fire on its
+   own — keep it only for a skill meant to be invoked by name alone.
+4. Replace this body with real steps, gotchas, and the expected output shape.
+   Ground it in this repo: concrete file paths, real commands, the non-obvious
+   traps specific to this codebase — not generic advice.