Agent Skills Manager — manage-skills Skill
Discover, create, edit, toggle, and migrate skills and rules across 11 AI coding tools — Cursor, Claude, Agents, and more — straight from the terminal, no scattered manual file juggling.
Purely documentation-based: teaches the agent to manage skill files via shell commands. Includes `rm -rf` deletion; deletion confirmation is explicitly required but there is no recovery mechanism beyond the rename-based .disabled toggle. Data flow is transparent: local file operations only, no network exfiltration. Publisher identity unverified; path accuracy rests on author claims. Deductions: unrecoverable deletes, confirmation only as prose guidance, limited sensitive-data handling notes.
Command snippets are self-consistent and use standard tools with 2>/dev/null tolerance; happy path is plausible. However there are no tests, thin edge-case handling (mkdir/cp will fail on missing directories), and some declared paths (e.g. ~/augment-guidelines.md, Windsurf directory layout) are independently unverified. Deductions: missing error-handling guidance, undefined behavior on abnormal input, no test evidence.
Target scenario is clear (unified management of skill files across tools), trigger phrases are enumerated in Guidelines, and both global and project scopes are covered. But boundaries and non-fit ranges are absent, and consistency of declared paths with current tool versions is unverified. Local-only operation poses no mainland-China network dependency. Deductions: missing boundary declarations and version-fit evidence.
Well-structured: path tables, format spec, per-operation sections, guidelines and naming conventions; repo carries MIT license. But the skill itself has no version number, no changelog, no FAQ, and only thin known-limitation disclosure (the single-file 'replace entire file' risk is noted but brief). Deductions: missing versioning/changelog, FAQ and troubleshooting gaps.
As a terminal-side skill-management reference it directly completes list/create/copy/toggle/delete tasks with usable commands and clear marginal value over installing a GUI. However, single-file editing replaces the whole file and may clobber existing config, and correctness depends on unverified path claims. Static review cannot verify execution; capped at 7 per anchors. Deductions: no representative-output verification, single-file overwrite risk.
Primary evidence is the in-repo path tables and snippets, locally auditable by users, but there is no test suite, no CI workflow, and no third-party execution evidence corroborating the tools' actual directory layouts. Fact/inference separation is mostly clean but coverage is thin. Deductions: lack of cross-source corroboration and reproducible verification coverage.
- Deletion uses rm -rf and is unrecoverable; back up skill files yourself before deleting — do not rely solely on the agent's prose confirmation.
- Editing single-file tools (Copilot, Codex, Aider, Augment) replaces the entire file and may clobber existing configuration; back up originals first.
- Declared tool directory paths are unverified and may not match current tool versions; verify local paths before first use.
- Publisher is not verified by the FollowSkills registry and identity is unknown; this is a static source review with no commands executed.
- Cross-tool copies require manual format adaptation; inspect the target file after copying.
What does this skill do, and when should you use it?
This skill comes from the umutbozdag/agent-skills-manager repository (MIT licensed) and is a pure Markdown guide that teaches an AI agent to manage skills/rules files for 11 tools directly in the terminal. It documents each tool's global and project paths, distinguishes directory-based from single-file formats, and provides concrete shell commands for listing, creating, enabling/disabling, copying, moving, deleting, and searching. Once installed, you can ask your agent things like “list all my skills across all tools” without any web UI.
The skill runs no code itself; it instructs the agent to: list skill directories per tool with ls/find; create skills in SKILL.md (YAML frontmatter) or plain .md format; enable/disable by renaming to .disabled; copy/move skills between tools or between global and project scope with cp/mv; search content with grep; and delete with rm -rf. It covers Agents, Cursor, Claude, Windsurf, Copilot, Codex, Cline, Aider, Continue, Roo Code, and Augment — 11 tools total.
- A developer using multiple AI coding tools (e.g., Cursor and Claude) who wants to copy a skill from one tool to another
- A terminal-heavy user whose skills are scattered across hidden directories and wants a single-command inventory
- A user who wants to temporarily disable a skill while preserving its content for later re-enabling
- A developer who wants to scope a global skill down to a specific project so it overrides the global one
- A user who does not want the repo's web dashboard and only needs terminal-based management
What are this skill's strengths and limitations?
- Covers 11 mainstream AI coding tools, including global/project scopes and the Cursor plugin cache path
- Pure Markdown guide with no script dependencies — works in any shell-capable agent environment
- All operations use standard shell commands, making behavior transparent and auditable
- Includes safety guidelines such as confirming before deletion and adapting naming when copying across formats
- Describes operations only at the documentation level — no automation scripts; execution quality depends on the agent
- No test suite or evidence of path validation against the latest versions of each tool
- The disable mechanism relies on renaming to .disabled, assuming each tool ignores that suffix (a documented assumption)
- Excludes the repo's web-dashboard features (editor, Git install, bulk actions, etc.)
How do you install this skill?
Copy the skill folder from the cloned repo to your tool of choice:
- Cursor: cp -r skills/manage-skills ~/.cursor/skills/manage-skills
- Claude: cp -r skills/manage-skills ~/.claude/skills/manage-skills
- Agents: cp -r skills/manage-skills ~/.agents/skills/manage-skills
(First run: git clone https://github.com/umutbozdag/agent-skills-manager.git)
How do you use this skill?
After installing, trigger it with natural language, e.g.: “List all my skills across all tools”, “Create a new skill called code-review for Cursor”, “Copy my deploy skill from Agents to Claude”, “Disable the pr-review skill”. The agent follows the path tables and command templates in the skill to run the appropriate shell operations.