Lobster Doctor

🦞 龙虾医生 — Workspace 健康管家。体检+清理+技能瘦身+Memory归档。说句话就能用。

Security Scan

The skill does what its name/description promise (workspace health, safe cleanup, and trimming skill descriptions) and asks for no external credentials, but it will read and can modify many local workspace files (including other skills' SKILL.md), so review backups and run dry‑runs before applying changes.

Purpose & Capability

Name/description (workspace health, cleaning, skill slimming, memory archive, cron audit) align with the provided code: scripts scan .openclaw workspace, sessions, skills directories, estimate token usage, generate reports, and implement a 'slim' operation that edits SKILL.md files. No unrelated credentials, binaries, or network services are required for core functionality.

Instruction Scope

SKILL.md promises preview-first, backups, and local-only operation. The code implements those behaviours (backup dir, preview/dry-run modes). However the instructions do not explicitly enumerate every path the code will read/modify: agent sessions under ~/.openclaw/agents, openclaw.json, skills directories (including a global node_modules path), and SKILL.md files will be parsed and (if apply chosen) modified. The weekly report code checks channels config and contains a Feishu helper stub (it currently returns False and avoids sending). Users should expect the skill to read potentially sensitive session files and to edit other skills' frontmatter when applying slimming.

Install Mechanism

No install spec is present; the package is instruction‑only in the registry metadata and relies on local Python (scripts are included). There are no downloads from external URLs or archive extraction steps. SKILL.md shows a clawhub install command, but no installer script is provided in the package.

Credentials

The skill requests no environment variables or credentials. It reads local config files (e.g., ~/.openclaw/openclaw.json and cron/jobs.json) which is coherent with generating reports and detecting enabled channels. No secret-taking environment variables are required by the code.

Persistence & Privilege

always:false (default). The skill can modify other skills' SKILL.md files (skill_slim.py 'apply' mode) and writes backups into a workspace .cleanup-backup directory. This is consistent with its purpose (slimming and cleanup) but grants it the ability to change other skills' metadata on disk — users should be aware and prefer dry-run/preview and check backups before applying.

Assessment

This package appears to implement what it says: local workspace auditing, safe cleanup (with backups), and a tool to shorten SKILL.md descriptions. Before using: 1) Run in report/dry-run mode only; do not run 'apply' until you've reviewed diffs. 2) Verify the backup folder (.cleanup-backup) is created and inspect backups. 3) Be aware it reads session files (~/.openclaw/agents/*/sessions/*.jsonl) which may contain chat history and sensitive content; if you have highly sensitive sessions, exclude or move them before running. 4) The slim operation can edit SKILL.md for many skills including globally installed ones (it scans an npm node_modules path); ensure you want those edits and have permission to restore from backups. 5) The weekly report contains a Feishu push helper stub that currently does not send; if that is later enabled, double-check channel configs. If you want lower risk: inspect the scripts locally, run tests (skill_slim_test.py) and limit the SKILL_DIRS before applying changes.

🦞 龙虾医生

你的 OpenClaw workspace 会越来越臃肿——会话历史堆积、临时文件堆积、技能描述膨胀。龙虾医生帮你自动治理。

核心功能

功能	一句话触发
体检诊断	「给龙虾做个体检」
安全清理	「帮我清理一下」
技能瘦身	「技能太多了」
Memory归档	「归档过期记忆」
会话健康	「会话健康怎么样」
cron巡检	「检查僵尸任务」

独家功能：技能瘦身

问题：每个技能的 description 都注入系统提示。100+ 技能 = 每轮白白烧掉几千 tokens。

解法：精简 description，只保留核心功能句。

效果：每轮节省 ~574 tokens。

安全机制

• 预览优先：任何修改都先展示预览
• 自动备份：清理前自动备份到 .cleanup-backup/
• 白名单保护：核心文件永不删除
• 零 API 调用：纯本地运行

安装

clawhub install lobster-doctor-skill-slim

更新日志

• v2.1.0: Memory归档命令、修复去重算法
• v2.0.0: 会话健康监控、周报生成

License

MIT