Released Agent Plugins CLI reference from the exact source. Exact source.

Released Agent Plugins CLI reference.

agentplugins author skills init

Atomically add one Skill to an absent skills/<name> destination

Synopsis

Atomically add one Skill to an absent skills/<name> destination

The package path is the exact standard root containing plugin.json; omission uses only the current directory. No ancestor search. Description is required explicitly for init; no prompts, inferred authors or identity normalization. Validate reports package-wide readiness and isolated immediate Skills; scripts, references, assets and allowed-tools are not execution or authorization evidence.

Installer-only inherited flags --scope, --accept-security-risk, and --security-details are rejected. Use the positional package path and author doctor for project checks; use agentplugins add for installation/security policy.

agentplugins author skills init <name> [package-path] [flags]

Examples

  agentplugins author skills init docs-helper ./package --description 'Use for documentation requests'
  agentplugins author skills validate ./package
  agentplugins author skills validate

Options

      --description string   required explicit Skill description (1–1024 characters)
  -h, --help                 help for init
      --include-root         disclose the explicitly selected package root

Options inherited from parent commands

      --accept-security-risk      continue despite blocking automated security findings
      --color auto|never|always   human output color: auto, never, always (default auto)
      --dry-run                   show the exact plan without changes
      --format string             output format: human or json (default "human")
      --no-color                  alias for --color=never
      --plain                     use accessible line prompts (independent of color)
      --scope string              installation scope (user only in this release) (default "user")
      --security-details          show every automated security finding in human output
      --target string             target client(s), comma-separated: codex, chatgpt, cursor, copilot, vscode, kiro, claude, gemini, opencode, cline, windsurf

SEE ALSO

• agentplugins author skills - Create and validate standard package Skills