编程

Skills Manager CLI

试用

Drive the Skills Manager CLI (`skm`) to initialize the hub, adopt unmanaged skills, list/enable/disable skills per AI tool, and doctor/fix symlink sync. Use whenever the user or an agent needs to manage skills from a terminal, SSH session, CI job, or headless machine; when a skill is missing in Claude Code, Codex, Cursor, Gemini, OpenCode, or any other supported tool; when asked to run skm / Skills Manager CLI; or when setting up Skills Manager without the GUI. Do not use for `npx skills` / skills.sh search (different CLI), for writing SKILL.md content (skill-creator), or for marketplace browse/install (not in skm).

它能做什么

Drive the Skills Manager CLI (`skm`) to initialize the hub, adopt unmanaged skills, list/enable/disable skills per AI tool, and doctor/fix symlink sync. Use whenever the user or an agent needs to manage skills from a terminal, SSH session, CI job, or headless machine; when a skill is missing in Claude Code, Codex, Cursor, Gemini, OpenCode, or any other supported tool; when asked to run skm / Skills Manager CLI; or when setting up Skills Manager without the GUI. Do not use for `npx skills` / skills.sh search (different CLI), for writing SKILL.md content (skill-creator), or for marketplace browse/install (not in skm).

技能文档

Skills Manager CLI (skm)

skm is the terminal front-end of Skills Manager. It reads and writes the same ~/.skills-manager/config.json and the same hub/tool symlinks as the desktop app, so GUI and CLI are interchangeable.

It does not create, edit, search, translate, tag, or marketplace-install skills. Those stay in the GUI (or you write files into the hub yourself, then skm enable).

Preconditions

  1. Resolve the binary. Prefer skm on PATH. If missing, say so and point at install (Settings → Command Line Tool, or a release archive skm-.tar.gz / .zip). Do not fall back to npx skills.
  2. Confirm with skm --version. A stale binary from an older app release should be updated from Settings or a matching release.
  3. Every command except init requires an initialized config. If stderr says so, run skm init --json once — it is idempotent and will not clobber an existing config. skm init and Settings → Install CLI also copy this companion skill into the hub and enable it for every currently active tool, so you usually do not need to install it by hand.

Hub path is fixed at ~/.skills-manager/skills (%USERPROFILE%\.skills-manager\skills on Windows). save() always rewrites skills_dir to that default; there is no supported way to pick another hub from the CLI.

Agent conventions

These exist because skm is designed to be called by agents, not just humans.

  • Prefer --json on init, list, doctor, fix, and adopt (apply). Data is one JSON document on stdout. Errors go to stderr as {"error":"..."} when the command was in JSON mode, otherwise error: .... A command can print a result on stdout and still exit 1 (fix with failed repairs) — parse stdout, then trust the exit code.
  • enable / disable have no --json. Success is a one-line enabled '' for '' (or disabled). Failure is error: ... on stderr, exit 1.
  • Never wait on a TTY prompt. adopt without --yes and without --json blocks on [y/N]. Pass --yes when applying. --json also skips the prompt (and therefore applies) — still pass --yes so the intent is obvious in the command line.
  • skm adopt --json --dry-run is not a machine-readable preview: if candidates exist it prints nothing. Preview with skm adopt --dry-run (human text), then apply with skm adopt --yes --json.
  • doctor --json exits 0 even when there are issues. Gate on .issues_count, not the exit code.
  • fix without --yes does not repair. JSON then looks like {"applied": false, "issues_found": N, ...}. Apply with skm fix --yes --json.
  • Write commands (init, adopt, enable, disable, fix --yes) take an advisory lock on config.json and fail fast if another skm is mid-write. Retry once. The desktop app does not take this lock.
  • Do not enable a skill for every detected tool unless the user asked. Pick the tool they named, or the host agent (see Which tool).

JSON shapes, exit codes, and the dry-run/apply matrix live in references/json.md. Read that before parsing output or writing a script.

Command map

IntentCommand
First-run / is it initialized?skm init --json
What skills exist, and where are they linked?skm list --json
Same, one tool onlyskm list --tool --json
Turn a hub skill on for a toolskm enable --for
Turn it off (remove the link, keep the hub copy)skm disable --for
Which tools are installed, any broken links?skm doctor --json
Repair what doctor reportedskm fix --yes --json
Preview leftover real dirs in tool skill foldersskm adopt --dry-run
Move those dirs into the hub and relinkskm adopt --yes --json

and accept an exact id, an exact instance_id, or a unique prefix (claudeclaude-code). Ambiguous prefixes error and list the matches — do not guess, rerun with a longer token. Exact tool id wins before prefix (trae is Trae, not Trae CN). Full ids and collision notes: references/tools.md.

Workflows

First-run on a machine with no GUI

command -v skm
skm init --json
skm adopt --dry-run
skm adopt --yes --json
skm list --json
skm doctor --json

init on an already-initialized config returns {"already_initialized": true, ...} and does nothing else. If config.json exists but cannot be parsed, init refuses to overwrite it — move or fix the file, then retry.

Make an existing hub skill available to a tool

skm list --json                          # find id / instance_id
skm enable  --for 
skm list --tool  --json            # confirm enabled_for

Enabled state is derived from the link on disk, not stored as a separate flag. enable creates the symlink (Windows: symlink → junction → tracked copy). disable removes that link; the hub copy stays.

If the same skill id exists in both global and the active project, a bare id is ambiguous. Use the instance_id from list: global: or project::.

Add a brand-new skill (CLI cannot scaffold it)

  1. Write ~/.skills-manager/skills//SKILL.md with YAML name + description. A directory is a skill if it contains SKILL.md, skill.md, or meta.json.
  2. skm enable --for
  3. skm list --tool --json to confirm.

Do not copy the folder into ~/.claude/skills (or any other tool dir) yourself — that bypasses the hub and adopt will later see it as an unmanaged real directory.

Adopt skills that already live in a tool directory

adopt looks at active tools (enabled + detected) for real directories that are not symlinks/junctions and not hidden. It moves each one into the hub, then puts a link back at the original path.

  • Duplicate hub names are skipped and left in place (contents are not merged — they may differ).
  • Preview first (--dry-run), then apply (--yes --json).
  • After apply, list should show the new ids as enabled for the source tool.

Diagnose "the skill is not showing up in Claude / Codex / …"

skm doctor --json
skm list --tool  --json

Read the results in this order:

  1. Binary missing or not on PATH → install, reopen the shell.
  2. config_initialized false → skm init --json.
  3. Tool detected: false → the tool's config dir is not on this machine. CLI does not re-detect existing tools on every call; a tool installed after init may still show not installed until the GUI redetects or the config is updated. Do not try to invent a custom tool from skm.
  4. Tool enabled: false → CLI cannot flip that flag. Use the GUI Tools page, or say so.
  5. Skill absent from list → it is not in the hub (and not in the active project). Write it into the hub or adopt.
  6. Skill present but not in enabled_forskm enable --for .
  7. link_issues > 0 or issues_count > 0 → skm fix --yes --json. If failed_count > 0, report the failed[].message values; do not loop fix.
  8. A wrong-target or deleted link is treated as disabled, not as a doctor issue. Re-enable to recreate the link over the tampered path.

fix only repairs what should_report_sync_issue flags. It will not delete a real (non-link) directory sitting in a tool's skills folder — that is what adopt is for.

Which tool

If the user named a tool, resolve it via references/tools.md and pass --for . If they did not:

Host--for
Claude Codeclaude-code
Codexcodex
Cursorcursor
Gemini CLIgemini
OpenCodeopencode
Unsureskm doctor --json, pick detected && enabled

Prefix shortcuts that are unique today: claudeclaude-code. Do not use trae when the user said Trae CN (trae-cn), or qoder when they said QoderWork CN (qoderwork-cn).

Out of scope (do not fake these with skm)

NeedWhat to do instead
Search / install from skills.shnpx skills / find-skills, or the GUI Marketplace
Author SKILL.mdWrite the file; this skill only covers linking it
Tags, notes, favorites, translationGUI
Add/remove a custom tool, bind a projectGUI
Enable or disable a tool itselfGUI Tools page
Change the hub directoryNot supported; always ~/.skills-manager/skills
Marketplace publish / ClawHubGUI

Install locations (when skm is missing)

The desktop app copies a bundled binary:

  • macOS / Linux: /usr/local/bin/skm if writable, else the first of ~/bin, ~/.local/bin that is already on PATH, else ~/.local/bin/skm (then the folder may need to be added to PATH)
  • Windows: next to the app executable (skm.exe)

Standalone: GitHub Releases skm-.tar.gz (.zip on Windows), extract, put on PATH.

相关技能

Claude Code skill management. Topics — writer (create), lint (validate + fix frontmatter), merge (combine related), dedup (find duplicates), convert (agent → skill), architecture (multi-topic structure), upgrade (enhance + add topics), route (topic placement), trigger (declare + auto-register hooks), find (discover via npx skills CLI), graph (extract depends-on + topic body Skill calls into Edge Table + Mermaid + dispatched d3 force-directed render), language (per-skill language consistency enforcement), portability (public skill cross-ref + vendor isolation), publish-scope (published skill scope review before extending), invoke-discipline (slash command → Skill tool call, multi-topic Read, post-decision auto-invoke, interactive script, vendor dispatch). Use when: "skill writer", "skill lint", "skill merge", "skill dedup", "create skill", "frontmatter fix", "multi-topic skill", "convert agent", "skill upgrade", "add topic", "topic route", "trigger compile", "hook auto register", "find

24 次安装

Audit and converge local skills across Codex, Claude, OpenClaw, OpenCode, workspace skills, and shared agent libraries; compute a hygiene score, classify sha...

16 次安装

Install skills from multiple skill hubs with one tool: clawhub.ai (default), skillhub.cn (SkillHub, China-optimized), skills.sh (npx skills, GitHub-based), t...

1 次安装

Expose installed agent Skills as HTTP(S) REST API services. Runs a persistent FastAPI server that auto-generates an API endpoint per Skill, with sync/async execution, webhook callbacks, and a multi-engine sub-agent executor (OpenClaw / Claude Code / Codex CLI / LLM fallback). Ships a bilingual (EN/ZH) web management console. HTTP by default (zero-friction), optional HTTPS with self-signed SAN certificates for production. Use when you need to serve Skills over HTTP, call Skills remotely, or expose Skill capabilities to external systems.