Claude Skills by VectorSpaceLab

Assemble validator-compatible reduced DIAYN recovery experiments that exercise fixed skills, discriminator rewards, and policy updates.

Build fixed-prior latent skill schedules and per-timestep conditioning records for DIAYN-style unsupervised rollouts.

Classify infrared-visible fusion approaches into explicit and implicit mechanism categories from textual method descriptions.

Compute deterministic no-reference proxy metrics for fused infrared-visible images, including entropy, contrast, edge preservation, and stability.

Fuse paired infrared and visible arrays while preserving thermal salience and visible texture through controllable local/global weights.

Run a bounded proxy recovery that compares explicit/implicit-style fusion and performs a small optimizer step on fusion weights.

Extract no-planning greedy policies from FB forward embeddings and a projected reward vector.

Create and validate reduced forward-backward successor-occupancy factorizations from reward-free transition tables.

Run a bounded soft-mode reduced recovery experiment for forward-backward representations with executable evidence.

Project late-specified rewards into backward representation coordinates for immediate FB policy adaptation.

Relabel downstream data with OPAL primitive latents and evaluate high-level latent control over temporally extended actions.

Prepare fixed-horizon offline state-action segments for OPAL primitive discovery and downstream latent relabeling.

Train or check an OPAL-style primitive autoencoding objective with reconstruction loss and KL-style prior matching.

Validate OPAL recovery evidence, mechanism checks, source boundaries, and proxy metric comparison.

Build a transferred greedy policy by maximizing action values over a library of prior policies or successor-feature heads.

Run a bounded successor-features plus GPI transfer recovery experiment for changing linear-reward RL tasks.

Compute and validate successor-feature reward decompositions for shared-dynamics reinforcement-learning tasks with changing linear rewards.

Select USFA transfer actions by generalized policy improvement over candidate policy encodings.

Compute linear multitask rewards, successor-feature values, and vector Bellman targets for USFA-style reinforcement-learning tasks.

Run a bounded Trip-MDP proxy experiment that validates USFA TD learning and GPI transfer mechanics.

Represent USFA policies with reward-weight encodings z and compute greedy policy actions under those encodings.

Rank simplified AutoAscend combat actions with symbolic safety, damage, and line-of-fire checks.

Select and interrupt prioritized symbolic NetHack strategies with explicit action queues.

Validate soft recovery traces by checking AutoAscend mechanism evidence and pass-rate metrics.

Build persistent symbolic NetHack state memory from compact observations for AutoAscend-style agents.

Apply AutoAscend nutrition and emergency resource rules to symbolic NetHack state memory.

Induce separated goal and effect rules from behavioral cloning traces while logging training evidence.

Evaluate GRAIL-style reactive behavioral clones with metrics, predictions, and mechanism checks.

Prepare fixed-rate behavioral cloning trace examples for GRAIL-style reduced recovery experiments.

Estimate and normalize diagonal Fisher importance vectors for Elastic Weight Consolidation from first-order gradients.

Compute Elastic Weight Consolidation's Fisher-weighted quadratic penalty, gradient, and additive multi-task quadratic terms.

Run a bounded EWC retention recovery comparison and emit validator-compatible result, trace, and mechanism-check artifacts.

Build deterministic sequential-task protocols for Elastic Weight Consolidation recovery experiments where old-task data is withheld after task switches.

Evaluate MAS in a bounded sequential-learning recovery and report forgetting plus mechanism checks.

Accumulate MAS importance across tasks and diagnose unlabeled subset adaptation.

Apply the MAS quadratic regularizer during later-task training and log parameter drift evidence.

Estimate Memory Aware Synapses parameter importance from unlabeled inputs using output-sensitivity gradients.

Apply NLE-style movement commands and task rewards for reduced staircase and score experiments.

Validate NLE full or reduced action-space choices and invalid-action penalties.

Run a deterministic reduced learning update over NLE-style symbolic features and rewards.

Normalize and validate NetHack Learning Environment symbolic observations for recovery harnesses.

Compute reduced NLE task rewards and clipping for symbolic recovery experiments.

Validate reduced NLE recovery outputs for metric metadata, mechanism checks, and source-boundary safety.

Run a tiny recurrent policy proxy with an optimizer step that mirrors NLE baseline training evidence.

Compute Random Network Distillation novelty bonuses and simple predictor updates for symbolic state features.

Parse NLE-style symbolic terminal observations into validated feature records for lightweight recovery experiments.

Compute separate intrinsic and extrinsic discounted returns for RND-style PPO with dual value heads.

Compute and test Random Network Distillation intrinsic rewards from predictor error against a fixed deterministic target feature map.

Maintain RND observation whitening/clipping and intrinsic reward scaling statistics.

Run a bounded soft-mode recovery experiment that validates core RND mechanism evidence without Atari-scale training.