196 skills · sorted by relevance
Guide reinforcement learning, training models through rewards, with slime. Prepare data, configure Megatron-LM training and SGLang response generation, and set up asynchronous runs or custom multi-turn workflows for supported language models.
Turn ePub books into training examples for models that reproduce an author's writing style. Covers text extraction, coherent segmentation, varied instructions, model training, and checks for copied phrases and style transfer to new scenarios.
Reviews recent Codex conversations to identify coding habits, strengths, and recurring challenges. Produces a prioritized growth report with practical next steps and relevant HackerNews reading, then delivers it to your Slack direct messages.
Read existing CARLA sensors to view live camera or lidar output, save recordings, or inspect a single reading. Display multiple sensors together and check ROS 2 topic names, delivery settings, and whether publishing is enabled.
Deploy language model APIs and run batch generation with vLLM. Configure GPU memory use, quantization—lower precision model storage—and distribution across GPUs, then test response times and monitor how many requests the server handles.
Locate conversation logs for Claude Code or Codex CLI and replace the latest assistant response with a fixed agreement message. A dry-run option previews the target without writing changes to the log.
Generate A2UI v0.8 surfaces for the OpenClaw Windows tray node, which renders A2UI natively in WinUI/XAML (not a WebView). Read this when pushing canvas.a2ui.push/canvas.a2ui.pushJSONL to a Windows node - covers the 18 supported components (Row, Column, List, Card, Tabs, Modal, Divider, Text, Image, Icon, Video, AudioPlayer, Button, CheckBox, TextField, DateTimeInput, MultipleChoice, Slider), the icon-name enum, the A2UIValue tagged union, JSONL envelope shapes, and action.context security / dataBinding rules. Catalog-strict: components outside this list render as the fallback Unknown placeholder.
The English summary is being prepared. Showing the original description.
数值格式清洗工具。读取 JSONL 结构化数据，按字段规则统一单位、数值格式、范围和空值表达，输出规整后的 JSONL 和统计文件。 当用户提到数值格式化、单位统一、千分位、科学计数法、空值填充、数值规整等需求时使用此skill。 即使用户没有明确说出"数值格式清洗"，只要任务涉及按字段规则整理数值表达，就应该使用此skill。 不负责文本去重、标点清洗、字段校验或非数值内容整理。
The English summary is being prepared. Showing the original description.
Bind addressable evidence IDs from `papers/evidence_bank.jsonl` to each subsection (H3), producing `outline/evidence_bindings.jsonl`.
The English summary is being prepared. Showing the original description.
Retrieve arXiv paper metadata with keyword queries or import an offline arXiv export, and save results as JSONL (`papers/papers_raw.jsonl`).
The English summary is being prepared. Showing the original description.
Claude Code session log schema — JSONL record types, message structure, tool call/result pairing, subagent file locations, team session layout, task/plan/team configuration paths. Use when parsing ~/.claude/projects/**/*.jsonl files, writing PostToolUse hooks that measure response sizes, building session analyzers, or any agent that reads or queries Claude session transcripts.
The English summary is being prepared. Showing the original description.
Capture outbound HTTP and telemetry from a CLI harness at runtime. Covers capture-channel selection (transcript file vs verbose-fetch stderr vs outbound proxy vs on-disk state), hook-driven per-event capture vs long-running session capture, JSONL output format for diff-friendly artifacts, and the observability table that maps each target to the cheapest channel that captures it. Use when a static finding needs runtime confirmation, when a payload shape is needed for a client re-implementation, or when dark-vs-live disambiguation requires watching what the binary actually sends.
The English summary is being prepared. Showing the original description.
Parse Claude Code JSONL session logs from ~/.claude/projects/ for tool call inventory, token costs, error detection, subagent traces, and compaction detection
The English summary is being prepared. Showing the original description.
Reads, searches and counts Claude Code session transcripts (the .jsonl files under ~/.claude/projects, subagents too) with the csift command-line tool, and ships hooks that re-inject compacted turns or record pending questions. Use when asked what a past or running session said, did, edited, spent or is doing now; when about to grep, jq or script over a transcript jsonl (a type=="user" filter overcounts the person's messages), count how many messages the person typed, tool calls or tokens, or claim something was never done or decided; to rebuild a lost file or plan from what a session wrote, extract pasted images, restore turns a compaction summary lost, find a subagent's own id when CLAUDE_CODE_SESSION_ID names the parent, wait on or message another session or subagent; or when csift prints "unknown label selector", "show needs an address", "looks like a session/agent id, not a project path" or "0 matches — a DEFINITIVE absence". Not for semantic or embedding search: csift matches regular expressions only.
The English summary is being prepared. Showing the original description.
Guided setup for sharing moflo's durable learnings through a git-tracked artifact — for a whole TEAM on one repo, OR for one person across several MACHINES (a team of one). Use when the user says "share learnings with my team", "commit our moflo memory", "sync memory across my laptop and desktop", "set up memory.team_artifact", "auto-export learnings on every PR", or "learnings.jsonl keeps conflicting". Installs a pre-commit hook so the artifact rides along in every commit. An existing learnings.jsonl moves to the conflict-free layout automatically. For many worktrees on ONE machine, use /memory-worktree.
The English summary is being prepared. Showing the original description.
Observe the harness itself — (WS1) measure context budget as a deterministic, trendable inventory (tokens per CLAUDE.md / skill / agent, lexical redundancy, per-CI cold-write totals; committed history.jsonl), (WS2) reconstruct the real invocation trace (which skills/agents/tools fired and how they nested) from session transcripts, and (WS-A) roll the per-run CI traces up into a committed weekly usage record (per-skill invocation counts; usage.jsonl) — the usage side of the dead-weight join. Use for "context budget", "how big is CLAUDE.md / our skills", "is the harness bloating", "how do our agent loops actually run", "which skills actually get used", "trace this session", "loop station".
The English summary is being prepared. Showing the original description.
Claude Code の過去セッション（transcript JSONL）・設定・運用パターンを cclens と同梱スクリプトで収集し、指定された分析観点に沿って傾向分析と改善案を出す。/session-insights で明示起動されたときのみ使用する。キーワードマッチによる自動起動はしない。起動時に分析観点の指定が必須。
The English summary is being prepared. Showing the original description.
Compare standardized metrics across multiple ego-benchmark-harness runs by recalculating them from tasks.jsonl and session JSONL, never from HTML reports. Covers overall aggregates (success rate/average score), Odysseys per-rubric pass rates, duration distributions (avg/p50/p90/max), turns/tools/cost, error and exception counts, per-task success matrices, helper distributions, and the share of DOM operations. Use whenever users ask to compare data or metrics across runs, calculate aggregate/Odysseys/duration/turn/error/helper metrics, inspect differences among runs, or compare multiple rwb/real-world-bench runs. Use pi-session-analyzer instead for deep forensic analysis of why an individual task was slow or failed.
The English summary is being prepared. Showing the original description.
アーキテクチャ知識グラフ(memory MCP / memory.jsonl)の初期作成。「知識グラフを初期化して」「memory グラフを作って」「アーキテクチャグラフを登録して」といった依頼、またはプロジェクト初期構築完了後の仕上げとして使用する。コードベースを調査し、モジュール構成と依存関係を entities / relations として登録する。
The English summary is being prepared. Showing the original description.
Build training datasets for LLM specialization from production data, frontier model distillation, and synthetic bootstrapping. Use when formatting production logs into SFT data, distilling from frontier APIs, or preparing data for fine-tuning. Covers JSONL formatting, data quality validation, deduplication, and train/eval splitting.
The English summary is being prepared. Showing the original description.
セッションログ（~/.claude/projects の jsonl）を横断分析する。集計モード（既定）は skill 発火・tool 頻度・軌道修正シグナル・compaction・未発火 skill を定量集計し、mining モードはユーザー発話を全件マイニングして訂正の規則化と繰り返し指示の既定化の提案を作る。手動起動専用。境界: 単一セッションの振り返りは update-inst、実データを見ない静的監査は instructions-audit。
The English summary is being prepared. Showing the original description.
Claude Codeセッションの効率を分析・スコアリングするWebツール。セッションログ（JSONL）を解析し、7観点100点満点で「無駄の有無」を評価。npx session-evaluatorで起動し、ブラウザUIでセッション選択→分析→改善提案まで実行可能。
The English summary is being prepared. Showing the original description.
MUST USE when asked to find, read, list, search, inspect, fetch, export, or reconstruct coding-agent sessions across Codex, Claude Code/Desktop, Aside, OpenCode, Senpi/pi, OpenClaw, Factory Droid, Amp, Gemini/Kimi/Qwen CLIs, Codebuff, Roo/Kilo/Cline, Kodu, Cursor CLI, Aider, Kiro, Goose, Hermes, Crush, Zed, or unknown local agent logs. Uses the repo-backed TypeScript coding-agent-sessions CLI for transcripts, session IDs, rollout JSONL, state SQLite, Claude histories, Aside messages.jsonl, OpenCode sessions, child linkage, cwd/model/time/token filters, archives, and cost clues.
The English summary is being prepared. Showing the original description.
Repeat candidate searches and comparisons while keeping an append-only decision log. Record evidence, compare results with earlier winners, and check replay, freshness, and approval requirements before moving to live trading or production actions.