rundatarun

Interrogate the user about a plan or design one question at a time, walking the decision tree branch by branch and recommending an answer for each, until every decision is resolved and understanding is shared. User-invoked only via explicit phrases like 'grill me', 'interrogate this plan', 'stress-test this design', or '/grill-me'. Never auto-fires. Distinct from EnterPlanMode (explores and proposes) and the premortem rule (runs after a rough decision); grill-me runs before a plan solidifies, to extract and resolve the open decisions.

Decodes an AI company past the marketing site — what they actually do, where the substance vs slideware divide is, and how they compare against 3 contextual alternatives + 1 contrarian build-internal option, ending with a take-the-meeting verdict and 2-3 prep questions sharp enough to expose the gap in 90 seconds. ALWAYS invoke when the user says '/dossier <company>', 'have you seen <company>', 'decode <company>', 'what does <company> actually do', 'should I take a meeting with <company>', 'is <company> any good', or pastes a company website and asks for a read. Always comparative — never produces a single-company writeup without 3 alternatives + 1 contrarian framed against the user's specific use-case lens. V1 SCOPE: AI/data/ML companies only — if the company isn't AI flavored, surface that and ask before proceeding. Refreshes existing dossiers in place (preserves prior verdict in 'Previous decode' section). Skeptical-analyst tone. Output to a per-company directory at ~/vault/Knowledge/AI-Research/Companies/

Compare LLM and media-model performance using Artificial Analysis (artificialanalysis.ai). ALWAYS invoke when the user asks to compare models, asks 'which model is best/fastest/cheapest', asks about model intelligence/coding/math benchmarks, asks about tokens-per-second / TTFT / latency, asks about model pricing, asks about Elo ratings for image/video/TTS models, or names two+ models and wants a head-to-head. Covers all major LLMs (OpenAI, Anthropic, Google, Meta, DeepSeek, etc.) plus text-to-image / image-editing / text-to-speech / text-to-video / image-to-video Elo leaderboards. Cached 1h, no rate limit concerns.