AndrewSmigaj/OpenLLMRI

cdd

Uncertainty assessment before implementation — identify what you know, what you're guessing, and what to verify

competitive-design

Generate 2-3 genuinely different approaches to a problem and compare them

devils-advocate

Challenge a design or approach — find real weaknesses, not performative objections

review-best-practices

Does the design follow established software engineering principles? Separation of concerns, coupling, error handling, API design.

review-consistency

Do all the parts agree with each other? Cross-document and internal consistency.

review-deliverability

Can phases ship independently with value? Are boundaries at natural integration points?

review-drift

Does the implementation match the design? What was built, dropped, or changed?

review-evolution

What's locked in by this design vs. what stays flexible? Cost of being wrong about the future?

review-interfaces

Are the boundaries between components clean, minimal, and well-defined?

review-onboarding

Could a newcomer (developer or new Claude conversation) understand this and implement from it?

review-risks

Failure mode analysis — what breaks, how do you detect it, what's the blast radius, how do you recover?

review-scope

Is everything earning its complexity? Sunk cost and simplicity check.

review-trace

Walk a concrete scenario end-to-end through the architecture — find gaps and friction

thorough-review

Fan-out all review skills via parallel agents, then synthesize findings into one report

analyze

Analyze a saved clustering schema — read data, reason about patterns, write reports

server

Start, stop, and check status of backend and frontend servers

temporal

Run temporal basin captures — single runs, paired batches, and verification

categorize

Categorize generated outputs for a probe session along output axes

pipeline

Check analysis pipeline state for an experiment and suggest next step

probe

Design a new probe experiment — interactive co-design of sentence set and probe guide