Loading...
Loading...
Found 29 Skills
Produce a long-form, shareable markdown writeup on whether Claude has regressed on this user's work. A bundled Python script scans `~/.claude/projects/`, computes every metric, and renders a markdown skeleton with tables already filled — in ~2.5s. Claude fills a dozen short narrative placeholders and saves. Writes `./cc-canary-<YYYY-MM-DD>.md` suitable for pasting into a GitHub issue or gist.
Maintain `*-skills` README standards and checklist-style roadmap docs through one canonical maintenance entrypoint. Use when a repo needs profile-aware README maintenance, checklist roadmap validation or migration, or a bounded audit-first doc workflow with Markdown and JSON reporting.
Analyze Huawei Ascend NPU profiling data to discover hidden performance anomalies and produce a detailed model architecture report reverse-engineered from profiling. Trigger on Ascend profiling traces, NPU bottlenecks, device idle gaps, host-device issues, kernel_details.csv / trace_view.json / op_summary / communication.json. Also trigger on "profiling", "step time", "device bubble", "underfeed", "host bound", "device bound", "AICPU", "wait anchor", "kernel gap", "Ascend performance", "model architecture", "layer structure", "forward pass", "model structure". Runs anomaly discovery (bubble detection, wait-anchor, AICPU exposure) alongside model architecture analysis (layer classification, per-layer sub-structure, communication pipeline). Outputs a separate Markdown architecture report alongside anomaly analysis.
Summarize deep research results into markdown report, cover all fields, skip uncertain values.
Must be used when users explicitly request "recommend submission journals", "help me choose SCI journals for my paper", "which journals is this manuscript suitable for", "journal matching/journal selection/submission suggestions". Applicable to scenarios where users provide full text, abstracts, Markdown, LaTeX, PDF, Word, or mixed materials; This skill will first use the built-in `2023IF.xlsx` to perform minimum hard filtering to generate a candidate pool based on the manuscript and user preferences, then the host model will independently plan Set1/Set2/Set3, verify the scope / quality / PubMed papers of the last 3 months via the internet, and finally output a Markdown journal selection report sorted by recommendation level. ⚠️ Not applicable: Users only want to polish papers, only want to translate abstracts, or only ask about the official website information of a single journal without needing systematic journal selection.
Interpret the meaning of paper figures and output a highly readable Markdown report that 'teaches humans how to read figures'; supports input of absolute paths to one or more figure files and manual interpretations, automatically attempts to retrieve the source code used to generate the figures from the vicinity of the figures, and uses a parallel-vibe-like approach to interpret each figure with process-level isolation via `codex exec`/`claude -p` (default concurrency limit is 3, adjustable in config.yaml). ⚠️ Not applicable: Users only want to adjust figure size/crop/change format; or request direct modification of images/source code (this skill has read-only access to images and source code throughout, modification is strictly prohibited).
Captures key decisions, questions, follow-ups, and learnings at end of a coding session. Writes a single markdown file per session. Use when done with a session, wrapping up work, running /done, creating a session summary, saving session context, or ending a coding session.
Local code review tool for self-inspection before git push. Triggered when users request phrases like "review my code", "check code changes", "review this commit", "review this", "code review", "git review", "help me check my code". Supports reviewing unstaged, staged uncommitted, and committed unpushed changes, and outputs a Markdown review report with scores.
Deep dive into a book. Collect information from six dimensions including chapter structure, background key points, problem impacts, solutions, term index, and further reading through parallel sub-agents, then output a Markdown deep learning note after cross-analysis. Trigger words: Analyze the book XX, Study XX, Reading notes for XX, book analysis.
Maintain JSONL-only profiler performance test cases under csrc/ops/<op>/test in ascend-kernel. Collect data using torch_npu.profiler (with fixed warmup=5 and active=5), aggregate the Total Time(us) from ASCEND_PROFILER_OUTPUT/op_statistic.csv, and output a unified Markdown comparison report (custom operator vs baseline) that includes a DType column. Do not generate perf_cases.json or *_profiler_results.json. Refer to examples/layer_norm_profiler_reference/ for the reference implementation.
Supports commands: [commit|commit_a commit_b] [--verify] [--apply] [--commit] Interactive git diff review skill. Parses git diff output into individual hunks, presents each hunk to the user with analysis for accept/reject decisions, verifies complete coverage, and generates a Markdown review report.
Gitコミットを日付ごとに集計してMarkdownテーブルで出力するスキル。 機能: (1) 指定日のコミット一覧取得 (2) worktree含む全ブランチ対応 (3) PR番号の自動取得 (4) JST時刻でのテーブル出力