← 返回 Skills 市场
Plan Mode
作者
pieterjanliekens
· GitHub ↗
· v3.0.4
· MIT-0
240
总下载
1
收藏
2
当前安装
11
版本数
在 OpenClaw 中安装
/install plan-mode
功能描述
Think-first execution with approval gating. Use when work is complex, ambiguous, irreversible, multi-step, worth comparing before choosing, interrupted and n...
安全使用建议
This skill appears coherent and low-risk: it instructs the agent to be read-only until you explicitly approve execution or persistence. Before installing, confirm runtime policies about network access and file writes: if you don't want the agent to query external registries (ClawHub/GitHub) or write plan files (e.g., docs/plans/), restrict network/file permissions or verify the agent will ask before performing those actions. Also review agent logs or audit prompts during the first uses to ensure plans are only persisted or actions performed after your explicit approval.
功能分析
Type: OpenClaw Skill
Name: plan-mode
Version: 3.0.4
The 'plan-mode' skill bundle is a structured framework designed to enforce a 'think-before-acting' workflow for AI agents. It includes explicit safety instructions in SKILL.md, such as maintaining a read-only stance until user approval, prohibiting the exfiltration of secrets or private data during tool audits, and requiring authorization before accessing external registries like ClawHub. The logic is focused on task management and risk mitigation rather than unauthorized execution or data theft.
能力评估
Purpose & Capability
The name, description, and SKILL.md consistently describe a read-first planning stance (clarify → plan → approve → execute) and the allowed actions (reads, listing, searching) match that purpose. No unrelated credentials, binaries, or installs are requested.
Instruction Scope
The runtime instructions explicitly limit the agent to read-only exploration until user approval, and specify what to do in each lens. The included references mention a 'toolbox audit' (scan installed skills, search ClawHub/GitHub) — which implies potential external registry or network lookups and checks of installed skills. This is reasonable for avoiding reinventing functionality, but it could cause the agent to perform network queries or inspect installed-skill metadata unless the runtime restricts that. The SKILL.md does not instruct any destructive actions prior to approval.
Install Mechanism
Instruction-only skill with no install spec and no code files — nothing is written to disk by an installer. Low install risk.
Credentials
No environment variables, credentials, or config paths are required. The skill's needs are proportional to its planning purpose.
Persistence & Privilege
The skill supports creating 'living' plan files but explicitly requires explicit user approval before persisting. It is not always-on. Autonomous invocation is allowed (platform default) but not combined with other concerning privileges in this skill.
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 plan-mode - 安装完成后，直接呼叫该 Skill 的名称或使用
/plan-mode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
v3.0.4
Clarify approval gating: no plan-file writes before explicit approval, toolbox audit is local-first, and external registry lookups require approval.
v3.0.3
Scrub context-specific example references from SKILL.md and replace them with neutral examples.
v3.0.2
Apply MiroFish review fixes: stronger auto-router decision rules, move non-use guidance to the top, define read-only more explicitly, add control semantics for off/cancel, improve living-plan resume behavior, add depth thresholds, tighten toolbox-audit scope, and add parallel/living examples.
v3.0.1
Tighten v3 based on first-pass MiroFish-style critique: clearer auto-router heuristics, stronger core-vs-advanced guidance, shorter description, explicit /plan off behavior, concrete living-plan convention, and failure-mode guardrails.
v3.0.0
Redesign as multi-lens planning system with clarify/compare/execute/recover/parallel/living modes, read-only-until-approval contract, depth modifiers, and stronger resume/surprise handling.
v2.1.0
v2.1: Complete rewrite for marketplace presentation. Human-friendly landing page with comparison table, concrete example, quick start, and toolbox audit highlight. Technical instructions restructured for scannability.
v2.0.1
v2.0.1: Updated installation docs to explain /plan native slash command registration.
v2.0.0
v2.0: Renamed skill to 'plan' so /plan registers as a native slash command. Restructured with features section, quick reference table, installation guide. Added /plan light mode. MIT license.
v1.2.0
v1.2: Restructured to match top ClawHub skills. Added: hook tagline, features section, quick reference table, installation section, multi-agent compatibility note, MIT license. Improved scannability and marketplace presentation.
v1.1.0
v1.1: Trimmed description for marketplace. Added /plan light mode. Sharpened activation criteria to reduce false triggers. Clarified living-document file path. Folded context awareness into Phase 2. Generalized benchmark stats in references. Addressed feedback from 4 independent reviews.
v1.0.0
Initial release: Research→Plan→Approve→Execute pattern with toolbox audit, living documents, and delegation support. 8 tools analyzed (Claude Code, Aider, Codex, Cursor, Windsurf, Roo Code, ExecPlan, claw-superpowers).
元数据
常见问题
Plan Mode 是什么？
Think-first execution with approval gating. Use when work is complex, ambiguous, irreversible, multi-step, worth comparing before choosing, interrupted and n... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 240 次。
如何安装 Plan Mode？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install plan-mode」即可一键安装，无需额外配置。
Plan Mode 是免费的吗？
是的，Plan Mode 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
Plan Mode 支持哪些平台？
Plan Mode 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 Plan Mode？
由 pieterjanliekens（@pieterjanliekens）开发并维护，当前版本 v3.0.4。
推荐 Skills