Invoking a skill
You don’t need a command or syntax. Describe what you need in plain language and Ultron routes to the right skill:
Run a competitive analysis on HubSpot.
Write a cold email to the VP of Engineering at Stripe referencing their recent infrastructure blog post.
Create a LinkedIn post about why most B2B cold emails fail.
Good vs. vague objectives
| Too vague | Specific |
|---|
| ”Write an email" | "Write a cold email to the Head of Growth at Linear referencing their recent Series B" |
| "Research a company" | "Deep dive on Notion — I’m pitching their CTO next Thursday" |
| "Create content" | "Write a LinkedIn post about why most SDR-to-AE handoffs fail, for a sales ops audience” |
What you see during execution
Tool activity streams in real-time as the skill runs:
[skill] competitive-analysis: starting
[tool] web_search("HubSpot pricing 2026")
[tool] web_search("HubSpot vs competitors features") ← parallel
[tool] scrape_website("hubspot.com/pricing") ← parallel
[tool] search_memories("competitor research hubspot")
[result] 4 sources processed
[tool] save_memory({ tags: ["cortex", "competitive", "hubspot"] })
[skill] competitive-analysis: complete
After execution
Every skill run automatically:
- Saves results to memory — tagged with skill category and the entity researched
- Logs activity — timestamped entry in the activity feed
- Returns the result inline — text, table, or canvas in the chat thread
You don’t need to save anything manually.
Cancelling a skill
Type “stop” or “cancel” at any point to interrupt a running skill. Partial results are still returned if any tool calls completed.
If a skill produces output that’s close but not quite right, follow up directly in chat: “Make the email shorter” or “Add a SWOT section to that analysis.” Ultron updates in the same thread.
