Agent skills for Claude Code, Codex and other agents: drive and record terminal apps, direct videos
FAQ (answers come from our records)
How do I install it?
The repository provides this install command: npx skills. It comes from the repository itself; we cannot guarantee it runs as-is in your client.
Who is it for?
We filed it under these tags: 开源项目. The tags come from the entry itself, not a taxonomy we imposed.
Who maintains it?
repository lorenzh/skills;first catalogued 2026-10-10;last seen 2026-10-16。There is no official certification or quality rating here — we record sources and dates only.
skills
以下摘录自该仓库的 README(片段) · 在 GitHub 读完整 README →
Agent skills for coding agents: Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI and any other agent that reads the open Agent Skills format.
A skill is a folder with a SKILL.md file. It teaches an agent one workflow, step by step, and can bring scripts and reference files with it. The agent loads a skill only when a task needs it.
Skills
| Category | Skill | What it does |
|---|---|---|
| research | researching-with-two-models | Has two researchers from different model families answer one brief, checks each disagreement against primary sources, and writes one report. |
| publishing | checking-before-publishing | Scans text, recordings and frames for private data (home paths, hosts, emails, tokens, internal names) before anything goes public. |
| agents | writing-skills | Authors skills with evidence, tiered checks and a bounded review budget. |
| agents | competing-implementations | Runs two implementations of one brief, compares blind diffs, and ships the reviewed winner. |
| agents | reviewing-across-families | Runs a review loop where a model of another family reviews an agent's change, the implementer fixes, and the orchestrating session ships the PR. |
| git | creating-worktrees | Sets up git worktrees that are really isolated: in-repo placement, dependencies installed and resolving inside the worktree, local files copied only with consent, and safe cleanup. |
| terminal | driving-terminal-apps | Drives, tests, screenshots and records terminal apps (CLIs, TUIs, REPLs) through tmux, like a person at the keyboard. |
| video | directing-videos | Plans, scripts and paces a video (demo, screencast, tutorial, launch video), then checks the finished cut against the plan. |
| publishing | listing-a-project | Finds active awesome lists and directories for a project, follows each one's submission rules, and submits only with the owner's consent per target. |
| agents | suggesting-skills | Reviews the current session for repeated workflows and hard-won fixes, checks them against existing skills, and offers a ranked list of new skills or updates. Creates nothing without consent. |
| video | editing-videos | Crops, zooms and captions recordings into per-beat clips, joins them with crossfades, and exports square, vertical or landscape MP4s with ffmpeg. |
| shipping | cutting-releases | Takes a release from version bump through tag, workflow and registry to a back-merge into main and a verified installed version, following the project's release document. |
| shipping | retargeting-stacked-prs | Checks whether a PR stack is ready, merges it as a native GitHub stack or retargets each PR onto main after its base merges, and gets fresh CI on every head. |
| shipping | waiting-for-ci | Waits on CI runs, PR checks, registries and background processes with commands that end on their own, bounded polls, state checks instead of guesses, and a diagnosis for failed checks. |
| shipping | landing-parallel-prs | Merges a batch of PRs that edit the same lines one at a time, updating each from main right before its merge, with the user's OK for every merge. |
| github | triaging-issues | Checks every open issue against the current code with cited evidence, plus milestone fit, type labels, priority and ordering conflicts; reports one table and changes nothing on GitHub until the user approves. |
| github | reproducing-bugs | Reproduces reported bugs of terminal apps in a real run on a clean checkout of the default branch, judges only text the app drew, and records reproduced, not reproduced or not run in the issue after a privacy check. |