Loading...
Loading...
Found 173 Skills
Optimize existing Triton kernels for NVIDIA TileIR backend on Blackwell GPUs (sm_100+). Adds TileIR-specific autotune configs: occupancy, num_ctas, TMA descriptors. Covers kernel classification (dot-related, norm-like, elementwise, reduction), type-specific transformations, and PTX-vs-TileIR benchmarking. Triggered by: "optimize for TileIR", "add TileIR configs", "Blackwell optimization", "TMA descriptors", "2CTA mode", "occupancy tuning". Kernels use standard `import triton`; TileIR activates via ENABLE_TILE=1 when nvtriton is installed.
BaoStock A-share Data Platform, free and open-source, supports queries for stock quotes, K-lines, financial data, industry classification, and index constituent stocks; used when users need to obtain A-share historical quotes, financial statements, trading calendars and other data
Pose classification using ST-GCN (Spatial Temporal Graph Convolutional Network). Classifies skeleton sequences into action categories from pose-keypoint data. Use when training, evaluating, exporting, or running inference for a TAO pose-classification model. Trigger phrases include "train pose classification", "skeleton action recognition", "ST-GCN", "keypoint sequence classifier".
Design taxonomy structure for categories, tags, or hierarchical classification. Supports flat, hierarchical, and faceted patterns.
Domain-specific keyword mappings for research report generation. Includes ROS2, AI/ML, and general engineering keywords for chapter classification and content mapping.
Core ML, Create ML, Vision framework, Natural Language framework, on-device ML integration. Use when user wants image classification, text analysis, object detection, sound classification, model optimization, or custom model integration. Covers Core ML vs Foundation Models decision.
Proof-driven exploitation with 4-level evidence system, bypass exhaustion protocol, mandatory evidence checklists, and strict EXPLOITED/POTENTIAL/FALSE_POSITIVE classification.
Write and audit Python code comments using antirez's 9-type taxonomy. Two modes - write (add/improve comments in code) and audit (classify and assess existing comments with structured report). Use when users request comment improvements, docstring additions, comment quality reviews, or documentation audits. Applies systematic comment classification with Python-specific mapping (docstrings, inline comments, type hints).
Use this skill when managing production incidents, designing on-call rotations, writing runbooks, conducting post-mortems, setting up status pages, or running war rooms. Triggers on incident response, incident commander, on-call schedule, pager escalation, runbook authoring, post-incident review, blameless retro, status page updates, war room coordination, severity classification, and any task requiring structured incident lifecycle management.
4-phase code review methodology: UNDERSTAND changes, VERIFY claims against code, ASSESS security/performance/architecture risks, DOCUMENT findings with severity classification. Use when reviewing pull requests, auditing code before release, evaluating external contributions, or pre-merge verification. Use for "review PR", "code review", "audit code", "check this PR", or "review my changes". Do NOT use for writing new code or implementing features.
Manage PR crises using classification, golden hour response, crisis statement templates (3C framework), and reputation recovery planning. Use this skill when the user faces negative media coverage, a viral complaint, product safety issues, executive misconduct, or any situation threatening brand reputation — even if they say 'we're getting destroyed on social media', 'draft a response to this article', 'how do we handle this PR disaster', or 'prepare for potential backlash'.
Evaluate source credibility using primary/secondary classification, internal/external criticism, triangulation, and misinformation detection. Use this skill when the user needs to assess whether information is trustworthy, evaluate research sources, fact-check claims, or detect misinformation — even if they say 'can I trust this source', 'is this real', 'how reliable is this data', or 'fact-check this for me'.