9 skills · sorted by relevance
Investigate why a PyTorch model behaves a certain way by changing its internal values with pyvene. The skill guides experiments that swap or remove activations, train interventions, compare outputs, and save or share intervention configurations.
Guides experiments that change internal values in PyTorch models to test what causes their behavior. Uses pyvene to swap or remove activations, locate influential components, train interventions, and save experiments for reuse.
Provides guidance for performing causal interventions on PyTorch models using pyvene's declarative intervention framework. Use when conducting causal tracing, activation patching, interchange intervention training, or testing causal hypotheses about model behavior.
The English summary is being prepared. Showing the original description.
Fornece orientação para realizar intervenções causais em modelos PyTorch usando o framework de intervenção declarativa do pyvene. Use ao conduzir rastreamento causal, activation patching, treinamento de intervenção de intercâmbio ou testar hipóteses causais sobre o comportamento do modelo.
The English summary is being prepared. Showing the original description.
Causal reasoning in historical analysis. Covers distinguishing immediate from underlying causes, tracing chains of causation across time scales, identifying unintended consequences, and applying counterfactual reasoning to assess causal significance. Use when analyzing why historical events occurred, evaluating the relative weight of causes, or assessing whether outcomes were inevitable or contingent.
The English summary is being prepared. Showing the original description.
Investigate errors and health in a running Convex app using plain-language questions. It summarizes relevant logs with counts and representative evidence, checks deployment timing, and provides dashboard links when possible for verification.
Create original relaxed black-pen graphics from concepts or visual references, including sparse illustrations, narrative scenes, animals, objects, abstract relationships, icons, and logo marks. Use when the user wants quick hand-drawn line energy, causal story staging, non-equilibrium proportion, shorthand forms, open contours, structural line drift, misregistered junctions, selective retracing, physical pen texture, or a transferable illustration system that must not collapse into clean vector minimalism or polished period-sketch drawing.
The English summary is being prepared. Showing the original description.
Trace bugs, CI failures, and unexpected behavior from reproduced evidence to a causal mechanism and governing contract. Distinguish product, architecture, test-oracle, and harness defects; investigate intermittent failures with controlled experiments and report unresolved evidence explicitly.
The English summary is being prepared. Showing the original description.
At the end of a long working session, produce a counterfactual trace note — a backward walk through the session that identifies the small number of load-bearing moments (observations, failures, accidents, decisions) without which the final outcome would not exist. Use this whenever the user asks to "close the session", "run the butterfly", "trace back", "what actually mattered here", asks for a session retrospective, or asks how the session arrived at its result. Also use it when a session ends with a significant outcome (a fix, a design decision, a shipped artifact) and the user wants a note capturing why it happened, not just what happened. Do NOT use for simple summaries of short sessions — this skill is for causal tracing, and it only earns its cost when the session was long enough to contain real forks.
The English summary is being prepared. Showing the original description.