Loading...
Loading...
Found 2,136 Skills
Convert single-node scripts to multi-node Slurm sbatch jobs and debug common multi-node failures. Covers srun-native vs uv run torch.distributed approaches, container setup, NCCL timeouts, OOM sizing for MoE models, and interactive allocation.
Establish the minimal causal explanation for one reproducible or directly observed failure by building and falsifying competing mechanisms. Use when the missing outcome is diagnosis rather than report validity, review, or correction delivery.
Reproduce and resolve an authorized browser issue using runtime evidence.
Diagnose a failure from evidence and verify the original scenario after a fix.
Investigate a specific Nx cache miss, artifact download or extraction failure, same-hash non-replay, or suspected unsafe cache replay. Use this skill whenever a user asks why an Nx task did not use the cache, why cache artifact restoration failed, why a cache hit restored stale or wrong output, or to compare two cached task executions. Use the nx-cloud-api skill for every Nx Cloud read. This skill supports API-only investigations, preserves strict hash-evidence limits, and requests local configuration only when the question requires it.
Diagnose Fireworks training and deployment issues, including stuck or failed jobs, error messages, poor quality, checkpoint and resume problems, LoRA adapter load mismatches, and deployments that are ready but not serving. Use for systematic read-only triage before retrying. Use configure for a new or approved retry run; use research when the starting method, data, or cookbook example is unclear.
Retrieve GitHub Actions logs with gh, analyze failures, and fix CI errors locally.
Debug inference clients that use an attached provider and its native endpoint, including hosted APIs and host-local Ollama, vLLM, SGLang, TRT-LLM, LM Studio, or NIM. Use for provider attachment, endpoint policy, credential substitution, topology, and migration from the removed managed inference endpoint. Trigger keywords - debug inference, managed inference endpoint, local inference, ollama, lm studio, vllm, sglang, trtllm, NIM, inference failing, model server unreachable, credential_endpoint_mismatch, host.openshell.internal.
Giải thích MỘT thứ (một file, hàm, tính năng, cơ chế, khái niệm, lỗi, hay hệ thống) theo cấu trúc cố định BẢY BƯỚC — Tên gọi → Nguồn gốc → Lý do tồn tại → Cơ chế hoạt động (+ sơ đồ hệ thống & code) → Trade-off → Giới hạn → Vị trí trong toàn bộ hệ thống — và CĂN THEO NGƯỜI NGHE (engineer, sếp/manager, designer, PM, sinh viên, bố mẹ, trẻ 5 tuổi…): cùng bảy bước, đổi độ sâu, từ vựng, phép so sánh, giọng. CHỨNG bước "Cơ chế hoạt động" bằng RUNTIME thật — chạy code với đầu vào cụ thể, thêm instrument/breakpoint (pdb/debugpy · node --inspect · print/log), quan sát state THẬT — thay vì đọc-rồi-đoán, rồi dọn sạch instrument. Gọi khi user nói "teach me", "giải thích", "dạy tôi", "cái này chạy thế nào", "sơ đồ hoá", "ELI5", "explain like I'm…", "giải thích như cho trẻ 5 tuổi", "nói dễ hiểu", "giải thích cho sếp/manager/designer/bố mẹ/vợ tôi…", "làm sao nói với sếp cái này", "/teach-me". KHÁC /onboard-codebase và /join-project (cả dự án → wiki): teach-me phạm vi MỘT thứ, sâu, có sơ đồ, chứng bằng chạy thật, không ghi wiki.
Use this skill when you need to establish the cause of a failure, regression, anomaly, flaky behavior, or unexpected result. Focus on diagnosis, not implementing the fix.
Use when app, service, or user-mode driver C++ coroutine code holds a thread-affine lock across suspension and later hangs or fails. Not for all coroutine crashes or choosing lock performance.
Diagnose a failure from its symptom — reproduce, classify the broken invariant, route via the repo AGENTS.md Task Router, trace the smallest root cause, and on request land a minimal fix behind a regression test. Framework knowledge via om-framework-context. Use for "why does this fail", "debug this", "fix this bug", "regression", "napraw błąd".