菜单
Create or review OpenClaw skills for the claw-config agent. Default mode drafts a new skill (proposal-only) with docs-first diagnosis, change control, and claw-config safety constraints; review mode audits an existing skill for correctness and safety.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | claw-config-skill-creator |
| description | Create or review OpenClaw skills for the claw-config agent. Default mode drafts a new skill (proposal-only) with docs-first diagnosis, change control, and claw-config safety constraints; review mode audits an existing skill for correctness and safety. |
| compatibility | OpenClaw |
| metadata | {"author":"claw-config","version":"1.1","openclaw":{"emoji":"🧱"}} |
A blueprint skill for creating claw-config-specific skills: conservative config operations with docs-first diagnosis, change control, and consistency guarantees.
/claw-config-skill-creator <new-skill-name>
Produces a complete PROPOSAL (file tree + full contents) for a new skill.
/claw-config-skill-creator review <skill-path-or-name>
Outputs a review report: issues, risks, and concrete edits.
/claw-config-skill-creator is either:
review, or<new-skill-name>references/HELP.md) and ask the user to restate.Must output:
Default is proposal-only (no writes).
Must output:
A claw-config skill must clearly separate:
bindings chooses which agent receives a message.channels.* gates whether replies happen (mentions/allowlists/topic overrides, etc.).Created skills should not embed a curated list of docs pages.
Instead, they must include a short Docs-first protocol:
A good skill teaches how to discover the right doc, not which doc to memorize.
references/ROLE_AND_BOUNDARIES.mdreferences/CHANGE_WORKFLOW.mdreferences/TOOLING_AND_SECURITY_MODEL.mdreferences/ARCHETYPES.mdreferences/CREATE_TEMPLATE.mdreferences/REVIEW_CHECKLIST.mdreferences/PUBLIC_HYGIENE.mdreferences/HELP.mdTelegram group/forum provisioning and management. USE WHEN: inviting bots to a group, promoting bots to admin, creating groups/forums, creating topics, sending messages as user or bot, checking bot permissions. DON'T use raw curl/Telegram API for these — this skill wraps safe Telethon (user) and Bot API (bot) scripts. Also use when: you need to add/remove/manage members in a Telegram group where you have admin rights. Triggers on: invite bot, add bot to group, promote bot, create topic, telegram group setup, forum setup. Safety: user API (MTProto) = high-risk, bot API = low-risk, never mix.
A cognitive protocol for safely managing and auditing OpenClaw application upgrades. Analyzes configuration-level risks (schema, defaults) and runtime-level behavioral shifts (routing, sessions, streaming) using semantic changelog analysis to prevent silent breaking changes.
Restore or migrate OpenClaw session history from a source session to a target session. Use when recovering lost context, migrating sessions after Telegram topic migration, or transferring conversation history between sessions. WARNING This is a HIGH-RISK operation that modifies session files directly.
OpenClaw configuration management safety protocol and change reporting. Use when modifying OpenClaw configuration that affects runtime behavior: (1) editing openclaw.json (bindings, channels, agents), (2) changing workspace prompt files (SOUL.md, AGENTS.md, IDENTITY.md), (3) restarting gateway or agents, (4) installing/removing skills or tools, or (5) any action that may impact message routing or agent behavior.
Safely simplify OpenClaw `bindings` (routing rules) while preserving behavior. Use when bindings are redundant (per-topic spam), when accounts are exclusive to a single agent, or when you need to reduce binding count without breaking routing/activation.
Cross-group posting via proxy bot. Route messages from a high-privileged agent (main) to a low-privileged bot account in another group/channel. Use message tool with accountId to specify which bot sends. Covers warm-up strategies for topics/threads and troubleshooting why proxy messages don't appear.