Skill Routing
How Glean auto-selects a Skill from your description
You'll be able to
- Explain how Glean picks a Skill automatically from its description
- Use Thinking mode to make routing more reliable
Read first
Day 3 Lab
- Write a request in plain language that should trigger one of your Skills.
- Send it without naming the Skill; note which Skill Glean routed to.
- Turn on Thinking mode and send the same request; compare the routing.
- If it routed wrong, tighten the description's "when to use" line and retry.
- Record which phrasings route correctly — that is your measurement.
Working Example: Routing by Description
Glean reads your request and matches it against each Skill's description to pick one automatically — no explicit name needed. Routing is most reliable in Thinking mode, where Glean reasons over the candidates. Verified against assistant_skills.
You type: "give me my standup for this week"
Glean compares against Skill descriptions:
• "Weekly status digest" → when user asks for a weekly status/standup ✓ match
• "Release Notes Drafter" → when user asks to draft release notes ✗
• "Meeting recap" → when user asks to summarize a meeting ✗
→ Routes to "Weekly status digest" (best description match)Routing keys off the *description*, not the instructions — so a precise "when to use" line is what makes auto-selection land. Measure it: try several phrasings and track which ones route to the Skill you intended.
Check for understanding
Why does the Skill description matter more than its instructions for routing?
Check yourself
1. What is a Glean Assistant Skill?
2. Who can invoke a Personal Skill?
3. When Glean auto-selects a Skill, what does it primarily match against?
4. Which gives you deterministic control over which Skill runs?
5. A Skill keeps firing on the wrong requests. What is the most likely fix?
In 10 seconds
“You don't always name a Skill. Glean reads your request, matches it against each Skill's description, and routes to the best fit — most reliably in Thinking mode.”
Explicit Invocation