Office-agent skill inventory & health check. Scans any skills directory, measures size and context-token footprint, detects duplicates, and classifies skills into used / protected / closeable / manual-review buckets. Reverse-dependency scanning locks every skill referenced by automations, hooks or plugins. Read-only by default: two disclosed, consent-gated write targets — --overrides --apply --yes (programmatic close on platforms like WorkBuddy; dry-run without --yes) and --set-lang (saves the report language preference to ~/.workbuddy/skill-inventory.json) — plus one disclosed backup artifact: a timestamped ~/.workbuddy/settings.json.bak. safety copy created before the close write, path printed to the user. Report language: auto (default; the Agent decides per conversation language) / zh / en; one-off override via --lang; JSON output is always English-keyed. No network access, no subprocesses. English docs: SKILL.en.md. 办公型 Agent 通用的技能库盘点与效能体检。扫描任意技能目录,统计数量/体积/上下文 token 占用, 识别重复与近似技能,并按「有使用记录/受保护/可关闭候选/需人工确认」四档给出保守建议;反向依赖 扫描会锁定被自动化/Hook/插件引用的技能。默认只读,两处写路径目标均需显式授权:--overrides --apply --yes(仅 WorkBuddy 等可关平台,缺 --yes 只做预览)与 --set-lang(把报告 语言偏好写入 ~/.workbuddy/skill-inventory.json);另有一份已披露的备份工件——关闭写入前在 settings.json 旁生成带时间戳的 settings.json.bak. 安全副本,路径会打印给用户。语言设置三选一: auto(默认,Agent 按对话语言 决定)/ zh / en;--lang 仅本次运行覆盖;--json 输出恒为英文键。无网络、无子进程。 英文文档见 SKILL.en.md。

Install

openclaw skills install @elisabeth15501/skill-inventory