菜单
Extract text from images and screenshots locally with mac-ocr, Apple's on-device Vision engine. Use when bulk-triaging screenshots or images (sorting Desktop or Downloads clutter), reading text out of a PNG or JPG, or any time image text needs extracting without spending vision tokens.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
Write pull request review comments and replies to review threads in the user's voice. Use when reviewing a PR or diff and producing comments, when replying to a bot thread (CodeRabbit, Codex), when triaging or applying review findings, or when staging review feedback in a pr-<number>-review-comments.md file for the user to proofread before posting.
The user's full writing system for prose they author — README and docs text, landing copy, PR and commit descriptions, Linear proposals, design docs, Slack and email messages. Use when drafting or rewriting any human-facing prose, when a draft reads like AI wrote it, when stripping AI slop from documentation, or when relaying research output into a human-facing surface. CLAUDE.md carries the compressed rules; this carries the reasoning, the before/after pairs, and the per-format shapes.
incident.io assistant for working with incidents. View incident details, post updates to incident channels, interact with the incident agent, and query structured incident data. Use whenever the user mentions an incident ID (e.g., "INC-123"), asks about an incident, or uses the /incident command. Trigger proactively when you see an incident reference.
Freshens the repos listed in mani.yaml - syncs each with upstream, runs local gates, fixes CI on the default branch, and rolls template releases out to copier children. Arguments are a space-separated subset of repo names from mani.yaml; default is every repo.
On-demand UX critique for terminal UIs (Bubble Tea, Ratatui, Ink, Textual). Combines hyperskills:tui-design's layout/keybinding/color principles with impeccable's critique rigor (cognitive load, personas, heuristic scoring, P0-P3 severity), adapted for terminals — no CSS, no browser, no WCAG contrast math. Use when reviewing or critiquing a TUI's design, auditing keybinding discoverability, or checking terminal portability (NO_COLOR, monochrome degradation, multiplexer/SSH behavior).
Capture and persist browser screenshots or GIFs for visual review of a web UI. Use when verifying a UI change visually, reviewing a page's design or layout, recording a multi-step interaction (dropdown, hover, form flow), toggling dark mode for a comparison shot, or when a screenshot needs to be written to disk rather than only shown inline.
| name | local-ocr |
| description | Extract text from images and screenshots locally with mac-ocr, Apple's on-device Vision engine. Use when bulk-triaging screenshots or images (sorting Desktop or Downloads clutter), reading text out of a PNG or JPG, or any time image text needs extracting without spending vision tokens. |
npx --yes mac-ocr <file> wraps Apple's Vision framework, the same on-device engine
behind Preview's copy-paste. It runs fully local, costs nothing, needs no API key,
and ships a prebuilt binary via npm — no Xcode or Swift toolchain. Despite the name
there is no brew tap; npx --yes mac-ocr is the working invocation.
Use it for bulk triage. Fall back to model-based OCR (the Read tool) only for images mac-ocr cannot parse — handwriting, heavy stylization — or when the task needs layout or semantic understanding beyond raw text.
The Bash tool's sandbox has an intermittent bug with literal spaced paths:
"Screenshot 2026-01-01 at 9.35.00 AM.png" fails to open even when the path is
correct, and dangerouslyDisableSandbox: true does not help.
Always cd into the directory and use a glob or a variable instead of typing the
literal spaced path:
cd ~/Desktop && for f in Screenshot*9.35*; do npx --yes mac-ocr "$f"; done