← 返回 Skills 市场
99
总下载
0
收藏
1
当前安装
7
版本数
在 OpenClaw 中安装
/install outclaw-style
功能描述
Learn the user's writing style per outreach channel (email, LinkedIn, Twitter/X, WhatsApp, Slack, SMS, Discord, Telegram — whatever is connected for this ten...
安全使用建议
Key things to check before installing or enabling this skill:
- Confirm connectors and prerequisites: Ask the platform/author which connectors (gog gmail, slack-mcp-server, whatsapp-mcp-ts, linkedin-cli, etc.) and local scripts (scripts/message_classifier.py, scripts/style_evaluator.py) must be present and whether the platform will provide them. The skill references these but declares none, so it may fail or behave unexpectedly.
- Clarify credential handling: The skill will need access to your message stores (email, Slack, WhatsApp, Telegram, Discord, etc.). Find out where OAuth tokens/credentials are stored, whether the platform will prompt you, and whether the skill will request any new credentials. Do not assume 'no env vars' means no secrets are used — connectors often require them.
- Understand auto-invocation/privacy: The skill can be auto-invoked silently by outclaw-plan to process a channel and will read outbound messages and write raw exports to kb/raw/ and learned prompts to ~/.openclaw/outclaw/styles/. If you do not want automatic scanning of message history, disable auto-invoke or require user confirmation for runs.
- Inspect persistence and storage protections: Raw outbound messages will be saved to disk. Ask where kb/raw/ and ~/.openclaw/ are located on your environment, who can read them, and whether they are encrypted. If sensitive, request that raw exports be avoided or that the skill redact PII before persisting.
- Resolve inconsistencies around user confirmation: The references include a 'User Confirmation' step (showing sample messages for approval) but SKILL.md also describes silent auto-invocation and forbids prompting for 'preferred tone'. Ask the author to clarify when the skill will present samples for review and when it will run silently.
- Request missing artifacts or run in a sandbox: The skill refers to scripts that are not bundled. Ask the author for those scripts or run the skill in a test/sandbox environment first to observe behavior and logs.
- If you proceed, prefer explicit opt-in: Only enable the skill for tenants/channels you explicitly permit, and consider disabling automatic background runs until you are satisfied with its data handling and required connectors.
能力评估
Purpose & Capability
The skill legitimately needs access to outbound messages across many channels and to persist learned prompts locally, which matches its description. However, it references many external CLIs/services (gog gmail, linkedin-cli, slack-mcp-server, whatsapp-mcp-ts, telegram-mcp, discord-mcp, XActions/xurl, mac_messages_mcp, bsky-mcp-server) and local helper scripts (scripts/message_classifier.py, scripts/style_evaluator.py) that are not declared in the skill's metadata (no required binaries, no install instructions, no required env vars). That mismatch (needs many connectors but declares none) is concerning and can lead to silent failures or unexpected behavior if the platform does not supply those connectors.
Instruction Scope
Runtime instructions explicitly tell the agent to collect outbound messages, classify them, write raw message dumps to kb/raw/style-<tenant>-<channel>-<ts>.jsonl and persist learned prompts in ~/.openclaw/outclaw/styles/<tenant>/<channel>_style.md. The skill can be auto-invoked silently by outclaw-plan to process a channel with no learned style, meaning it may read and persist private messages without a direct user prompt. There is also an internal inconsistency: the references describe a 'User Confirmation' step presenting sample messages for approval, while the SKILL.md emphasizes silent auto-invocation and forbids soliciting 'preferred tone' — it's unclear when/if user confirmation happens.
Install Mechanism
This is an instruction-only skill with no install spec (lowest install risk). However, it depends on external CLIs and local scripts that are not included in the package. The absence of an install mechanism or bundled tooling means the skill will either fail when those commands are missing or rely on platform-provided connectors; that operational gap is important to verify before use.
Credentials
The skill requests no environment variables or explicit credentials in metadata, yet its operation requires access to message stores and connectors that typically need OAuth tokens/credentials. It also accesses local config/memory files (capabilities/<tenant>.json, tenant memory, kb, and home directory paths). The lack of declared credential requirements reduces transparency about what secrets will be used or required by the platform for it to function.
Persistence & Privilege
always:false (good). The skill will persist data to the user's home (~/.openclaw/outclaw/styles/) and to kb/raw/ (raw message exports). Persisting training data and learned prompts is consistent with the skill's purpose, but combined with auto-invocation (silent, one-channel runs triggered by outclaw-plan) this creates a privacy surface: private outbound messages will be written to disk without an explicit permission prompt unless the platform enforces such prompts. The skill does not attempt to modify other skills or system settings.
如何使用
- 确保已安装 OpenClaw（本地或 Docker 部署）
- 在对话框中输入安装命令：
/install outclaw-style - 安装完成后，直接呼叫该 Skill 的名称或使用
/outclaw-style触发 - 根据 Skill 的参数说明提供必要输入，即可获得结构化输出
版本历史
v1.0.6
v2.1.33: hard sanity gate refuses truncated Leadbay pulls — caps agent's paraphrase-and-summarize failure mode
v1.0.5
v2.1.32: ensure_leadbay_bound self-heal for plugins.allow + entries + agent skills + extensions dir
v1.0.4
v2.1.31: pull count=50, kill stale --n 3, heredoc-write discipline
v1.0.3
v2.1.30: paste-the-file-verbatim rule + sanity gate kills 'see above' bypass
v1.0.2
v2.1.29: first-action scaffolder rule + chat-reply-must-validate + angle-fit detector (kills 'furthers their sustainability goals' class)
v1.0.1
v2.1.28: agent mindset + options-and-choose + coherence checklist + evidence card + per-tenant style + daily launchd cron
v1.0.0
- Initial release of outclaw-style.
- Learns and saves per-channel writing style prompts for each tenant based on outbound message samples.
- Implements a 5-iteration Prompt Learning Protocol for style extraction and prompt refinement.
- Supports training via explicit user command or silent auto-invocation by outclaw-plan when needed.
- Stores learned prompts and related metadata in a resolver-compliant format, with strict consent and privacy handling.
- Generates compact user reports on training results, including best score and sample counts per channel.
元数据
常见问题
Outclaw Style 是什么？
Learn the user's writing style per outreach channel (email, LinkedIn, Twitter/X, WhatsApp, Slack, SMS, Discord, Telegram — whatever is connected for this ten... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 99 次。
如何安装 Outclaw Style？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install outclaw-style」即可一键安装，无需额外配置。
Outclaw Style 是免费的吗？
是的，Outclaw Style 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
Outclaw Style 支持哪些平台？
Outclaw Style 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（cross-platform）。
谁开发了 Outclaw Style？
由 milstan（@milstan）开发并维护，当前版本 v1.0.6。
推荐 Skills