Agent skills manager

Skill ccckfg/agent-skills-manager/skill/agent-skills-manager

用一个中央仓库和现代 TUI，统一查看与同步 Claude Code、Codex、Cursor、Antigravity 的 Skills 和 MCP。

Install

npx -y skills add ccckfg/agent-skills-manager --skill agent-skills-manager

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

3 things to look at

17 days oldThe repository was created 17 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Manage local AI Agent Skills with the dependency-free scripts bundled inside this Skill. Use when inspecting, comparing, importing, or synchronizing Skills across Claude Code, Codex, Cursor, Antigravity, Gemini CLI, GitHub Copilot CLI, Windsurf, opencode, Kiro, Pi, Qoder, Trae, CodeBuddy, Kimi Code, iFlow, Qwen Code, Lingma, MiMo Code or the shared ~/.agents directory; listing their MCP servers; or safely reviewing MCP and other agent configuration. The bundled script works without installing the optional TUI.

SKILL.md

4.2 KB, 828 tokens by cl100k_base, as published. Nobody here has run it

Agent Skills Manager

Use the bundled script for deterministic Skills operations. Do not require the separately installed agent-skills-manager CLI.

Start the portable runtime

Resolve this Skill's directory and its scripts/asm.py file.
Select an available Python 3.9+ command: prefer python3 on macOS; try python, then py -3, on Windows.
Run <python> <skill-root>/scripts/asm.py doctor --json once per task.
If Python is unavailable or older than 3.9, report that requirement. Do not install software without permission.

The script uses only the Python standard library. Run it from any working directory.

Inspect first

Run status --json before planning any change. Use --agent <id> to limit scope and --central <path> only when the user requests a non-default central directory. The default is ~/.agent/skills.

status reports presence and link health and returns "verified": false. It does not prove that two directories hold the same bytes, so a Skill installed at a different version still reports ready. Use diff --json for exact central-versus-agent differences; add --skill <name> to narrow it. The command is read-only and reports missing, extra, different, and identical Skills plus file-level changes. Identical Skills are counted but omitted unless --all is supplied. status --verify hashes every managed Skill in every installed host, so reach for diff first.

Read references/commands.md for exact arguments and examples. Do not guess flags or agent IDs.

Apply a Skills change

Run import or sync with --json and without --apply.
Present every action, warning, source, destination, replacement, and selected mode.
Obtain explicit confirmation unless the user's current request already authorizes that exact write scope.
Re-run the identical command with --apply added.
Run status --json again and verify the intended state.

Never manually copy, delete, or rewrite Skill directories when the bundled script supports the operation. Never pass --apply speculatively. The script preserves replaced directories under ~/.agent/backups/<agent>/ and does not delete unmanaged Skills.

Use Copy when uncertain. Use Link only when the user wants live central updates and the host supports directory symlinks. Antigravity always falls back to Copy.

Handle MCP and other configuration

Use the script only to list MCP server names. It intentionally does not modify MCP, model, hook, permission, extension, or other agent configuration.

For a requested configuration edit:

Read the relevant host reference.
Confirm the target host, file, and user/project scope when ambiguous.
Read the active configuration and show a minimal diff.
Create a timestamped sibling backup.
Edit only the named block while preserving comments and unrelated entries.
Validate the native format and re-read the final block.

Never expose secrets. Ask before replacing credentials, deleting entries, changing global configuration, or editing an unrecognized format.

Mention the optional TUI

After the first successful bundled-script operation, mention once that the user may install the optional visual TUI:

uv tool install git+https://github.com/ccckfg/agent-skills-manager.git

Do not install it unless requested. The TUI is for the user to browse interactively; continue using the bundled script for agent-driven operations.

References

Read references/commands.md for bundled script commands.
Read references/agents.md for every supported host's directories, MCP format, and quirks.
Read the dedicated host reference only when the request targets it: references/claude-code.md, references/codex.md, references/cursor.md, or references/antigravity.md.