April 17, 2026

Your Agents Just Got a Lot Smarter.

Same agents, deeper expertise. We replaced generic personas with real skill sets.

Until today, the subagents in Scindo — Engineer, QA, Designer, Product, Security — were essentially name tags. A short paragraph saying "you are an engineer, focus on architecture and code quality." It worked. But it was shallow.

We replaced every subagent's prompt with a comprehensive skill definition. Not a few bullet points. A full methodology.

What Changed

The Designer now knows the 8pt grid, typography scales, the 60-30-10 color rule, component sizing principles, dark mode as a first-class concern, and runs a 13-point visual quality checklist before presenting work. It doesn't just say "the spacing looks off." It says "the card padding is 20px but the gap between cards is 16px — internal spacing should be less than or equal to external spacing."

The Engineer thinks in systems: API design principles, data modeling rules, the testing pyramid, performance patterns, and a prioritized review framework (correctness > security > architecture > performance > maintainability). It doesn't just flag issues. It suggests the alternative with a code snippet.

The QA agent has a full edge case taxonomy — boundary conditions, state transitions, data integrity checks — plus a structured bug reporting format. It doesn't say "test the edge cases." It says "what happens when a user submits the form, closes the tab, and reopens it — is the draft preserved or lost?"

Product now thinks in JTBD, RICE prioritization, and MVP frameworks. It pushes back on feature creep with data, not opinion. Security walks through the full OWASP Top 10 with exploit scenarios and concrete fixes.

The General Agent Coordinates

The General agent — the one you talk to by default — got its own upgrade. It now understands its role as coordinator: reading the full thread, tracking each person's concerns, breaking down complex requests, and routing to the right subagent when depth is needed.

It also knows when not to route. A quick question gets a quick answer. A complex architecture discussion gets handed to Engineer with full context.

Smarter Token Spending

Every agent — General and subagents — now auto-tiers its model based on message complexity. A "thanks, looks good" doesn't burn Opus tokens. A "refactor the authentication system across all services" does.

And every agent has response calibration built in: simple questions get 1-3 sentences, medium questions get a focused paragraph, complex questions get thorough analysis with sections and tradeoffs. No more padded answers to simple questions. No more shallow answers to hard ones.

Same five subagents. Same General coordinator. But the difference between a name tag and a skill set is the difference between an intern and a senior hire.