Agent Skills by kangoxford
Showing 35 vetted skills indexed across 1 GitHub repositories.
checkstate
Reconstruct experiment narratives from git history and logs into status reports.
multi-node-jax-ffi-debug
A structured guide for deploying and debugging multi-node JAX applications, emphasizing GPU cluster-specific considerations.
planning-with-files
Create and maintain task_plan.md, findings.md, and progress.md for persistent planning.
symlink-claude-config
Convert a copy-based Claude configuration directory into a live symlink view of ~/.claude.
chmod-group
Adjust Unix group permissions on directory trees with recursive and path-only modes.
bench-mars
Benchmark MarS order-level checkpoints with GPU inference and LOBbench scoring.
siso-cuda-kernel-opt-tradeoffs
Analyze SISO CUDA kernel tradeoffs for state-space kernels across d_state values.
notion-idempotent-sync
Archive Notion page blocks and rewrite them from Markdown sources.
live-overleaf
Automate LaTeX drafting and push updates to Overleaf per topic.
list-recent-sessions
List recent Claude Code sessions with metadata in an ASCII table.
chmod-world
Configure world-accessible shared folders on HPC Lustre with cross-group read/write access.
moveon
Restore and revalidate SLURM job monitors after session reconnect.
lustre-safe-compress
Compress Lustre files with per-file verification and multi-pass quota retries.
reuse-checkpoint-for-ablation
Search SLURM logs and W&B configs for matching training checkpoints.
find-wandb
Locate WandB run URLs, directories, and configurations for SLURM jobs.
finding-lobbench-results
Locate existing LOBbench results across pipeline directories for a given checkpoint, job, run, or experiment.
notion-push-via-rest
Convert Markdown blocks into Notion blocks and create child pages via REST API.
notion-auto-sync-md
Synchronize Markdown file edits to Notion pages via PostToolUse hooks.
lustre-storage-audit
Audit Lustre storage usage by reconciling du results with lfs quota.
meta-learning-evolution
Automate meta-learning pipeline design by evolving code, prompts, and configurations.
eval-before-optimize
Validate whether evaluation improvements are statistically meaningful before optimizing post-training models.
fireworks-tech-graph
Convert natural-language system descriptions into SVG and PNG diagrams.
claudeception
Detect recurring patterns in completed sessions and generate reusable SKILL.md entries.
find-session-id
Identify the origin Claude Code session for a quoted snippet from JSONL logs.