Loading...
Loading...
Found 172 Skills
Adds PyTorch FSDP2 (fully_shard) to training scripts with correct init, sharding, mixed precision/offload config, and distributed checkpointing. Use when models exceed single-GPU memory or when you need DTensor-based sharding with DeviceMesh.
Migrates Nango syncs from deleteRecordsFromPreviousExecutions()/trackDeletes to trackDeletesStart/trackDeletesEnd for automated deletion detection (including checkpoint-based full refresh). Use when updating existing createSync code.
Orchestrates multi-day execution of complex tasks through milestones. Each milestone goes through plan-crafting, run-plan (worker-validator), and review-work phases with checkpoint/recovery. Triggers when the user says "long run", "start long run", "execute milestones", or "run all milestones".
Build timeout-resistant Claude Code workflows with chunking strategies, checkpoint patterns, progress tracking, and resume mechanisms to handle 2-minute tool timeouts and ensure reliable completion of long-running operations.
[Tooling & Meta] Restore workflow context from checkpoint after session loss
Use this skill when the user wants to interact with remote Sprites from their local machine — listing sprites, executing commands, managing checkpoints, transferring files, controlling network policy, or coordinating work across multiple sprites.
Workflow Checkpoint Basic Capabilities (Focus on Save and Resume): Record checkpoint progress and resume context in GitHub Issues. Applicable to any workflow stage, supporting automatic triggering and high-frequency manual calls. Keywords: save, resume, checkpoint, issue.
Meta-cognitive decision support that analyzes current context and surfaces intelligent next-step options to the user. Use this skill when: (1) User explicitly invokes /checkpoint, (2) Significant work has been completed and a checkpoint is valuable, (3) Uncertainty or ambiguity exists about requirements or approach, (4) Task complexity has expanded beyond initial scope, (5) Before finalizing or committing to ensure nothing is missed. This skill pauses execution, assesses the situation holistically, and presents 2-5 contextually-appropriate options via AskUserQuestion, with a recommended option and rationale.
Orchestrate the full paper pipeline end-to-end. Manage state propagation between phases (literature → plan → code → experiments → figures → tables → writing → review), support checkpointing and resumption. Use for assembling a complete paper from components.
Use when you need to execute I2 (Implementation Execution) in the Spec Pack of sdlc-dev, using `{FEATURE_DIR}/implementation/plan.md` as the sole SSOT to implement in batches, run minimal validation, write back audit information, and report at batch checkpoints; stop immediately when encountering blocks or clarification items.
Run this repo’s Units+Checkpoints research pipelines end-to-end (survey/综述/review/调研/教程/系统综述/审稿), with workspaces + checkpoints. **Trigger**: run pipeline, kickoff, 继续执行, 自动跑, 写一篇, survey/综述/review/调研/教程/系统综述/审稿. **Use when**: 用户希望端到端跑流程（创建 `workspaces/<name>/`、生成/执行 `UNITS.csv`、遇到 HUMAN checkpoint 停下等待）。 **Skip if**: 用户明确要手工逐条执行（用 `unit-executor`），或你不应自动推进到 prose 阶段。 **Network**: depends on selected pipeline (arXiv/PDF/citation verification may need network; offline import supported where available). **Guardrail**: 必须尊重 checkpoints（无 Approve 不写 prose）；遇到 HUMAN 单元必须停下等待；禁止在 repo root 创建 workspace 工件。
Controls InnerClaude instances on Sprites.dev VMs for testing workflows, install patterns, and Claude-to-Claude interaction. INVOKE BEFORE any 'sprite exec', 'inner Claude', 'test this workflow', 'Claude controlling Claude', or remote VM operations. Documents the critical tmux+pipe-pane pattern that makes OuterClaude/InnerClaude interaction work. Also covers checkpoint/restore and bootstrap. (user)