Menu
This skill should be used when the user wants to sketch content from a URL, YouTube video, or external source. Enables fetching and extracting content before creating a visual sketch.
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Basé sur la classification professionnelle SOC
Review and clean up commits before pushing. Uses planner/executor split: reads commits inline, authors final messages in saved style, replays via `git rebase --exec` with GIT_EDITOR=true so no editor opens. Detects logical clusters of granular TDD commits and proposes collapsing them into one logical commit. Auto-amends drifted subjects on revalidation. Activates on: "review commits", "clean up commits", "prep for push", "squash these commits", "finalize commits", "merge branch to main", "clean up my commits before pushing", "review before push".
Validate commits before pushing. Runs six deterministic checks: clean worktree, tests pass, no AI co-author leaks (Claude, Anthropic, GPT, OpenAI, Copilot), no conflict markers, no squash/fixup residue, and subject lines conform to the configured style (classic or conventional). Activates on: "validate commits", "check commits before push", "any AI leaks", "check for co-author", "are my commits clean", "validate before pushing", "check commits".
Triggers when user wants to PERFORM a git commit action. Activates on phrases like "commit these changes", "commit this", "let's commit", "make a commit", "create a commit", "commit my work", "commit the changes", "time to commit", "ready to commit", "commit what I've done", "save my changes" (git context), "commit for me", "do the commit", "commit now", "go ahead and commit", "please commit", "can you commit", "would you commit"
Provides educational guidance on writing good git commit messages. Activates when users ASK QUESTIONS about commit message formats, styles, conventional commits, classic commits, atomic commit principles, or git best practices. Does NOT activate when users want to PERFORM a commit (use the commit-action skill instead).
Initialize and manage Nix flake development environments with auto-detection and security tooling
Guardrail for efficient refactoring. IDEs use AST-based semantic refactoring (instant, atomic, accurate). AI uses text-based pattern matching (slower, sequential, error-prone for structural changes). This skill provides decision criteria for when to delegate to IDE vs proceed with AI. Consult before: renames across files, moving packages, changing signatures, extracting interfaces, or any multi-file coordinated structural change.
| name | capture-for-sketch |
| description | This skill should be used when the user wants to sketch content from a URL, YouTube video, or external source. Enables fetching and extracting content before creating a visual sketch. |
| version | 1.0.0 |
When the user wants to sketch external content (URLs, videos, articles), first fetch and extract the content, then generate the visualization.
Use this skill when the user mentions:
Step 1: Fetch video content
Use yt-dlp to get the transcript (if available):
yt-dlp --write-auto-sub --sub-lang en --skip-download -o "%(title)s" "<URL>"
Or fetch video metadata:
yt-dlp --print title --print description --skip-download "<URL>"
Step 2: Extract key information
Step 3: Generate diagram
Step 1: Fetch content
Use the WebFetch tool to retrieve and process the page:
WebFetch(url, prompt="Extract the main topics, key arguments, and structure of this article")
Step 2: Extract key information
Step 3: Generate diagram
Step 1: Fetch README
WebFetch("https://github.com/{owner}/{repo}", prompt="Extract the architecture, main features, and components")
Or if repo is local, read the README.md directly.
Step 2: Extract key information
Step 3: Generate diagram
Title: {video title}
Main Topic: {core subject}
Key Points:
1. {point 1}
2. {point 2}
3. {point 3}
Flow: {step1} → {step2} → {step3}
Takeaways:
- {takeaway 1}
- {takeaway 2}
Title: {article title}
Source: {URL}
Main Argument: {thesis}
Supporting Points:
- {point 1}
- {point 2}
Key Concepts:
- {concept 1} relates to {concept 2}
Conclusions: {summary}
Project: {repo name}
Purpose: {description}
Components:
- {component 1}: {description}
- {component 2}: {description}
Dependencies:
- {component 1} → {component 2}
Tech Stack: {technologies}
User: "/sketch https://youtube.com/watch?v=abc123"
yt-dlp --print title --print description --skip-download "https://youtube.com/watch?v=abc123"
${workbench_path}/sketches/User: "/sketch https://example.com/article-about-microservices"
WebFetch(url, "Extract main topics and structure")
${workbench_path}/sketches/User: "/sketch https://github.com/facebook/react"
${workbench_path}/sketches/Before using yt-dlp, check if it's available:
command -v yt-dlp >/dev/null 2>&1
If not available, offer alternatives:
If content extraction fails:
Sketches are saved to ${workbench_path}/sketches/ (shared with jot plugin).
Read workbench_path from .claude/jot.local.md:
---
workbench_path: ~/workbench
---
Default: ~/workbench if not configured.