← 返回 Skills 市场
829
总下载
2
收藏
3
当前安装
2
版本数
在 OpenClaw 中安装
/install close-loop
功能描述
End-of-session workflow for shipping changes, consolidating memory, applying self-improvements, and preparing publishable outputs with safety gates.
使用说明 (SKILL.md)
Close Loop
Use this skill when the user says "wrap up", "close session", "end session", "close out this task", or invokes /wrap-up.
Run four phases in order and return one consolidated inline report.
Modular structure
This skill is split into components for maintainability:
components/01-design-principles.mdcomponents/02-phase-1-ship-state.mdcomponents/03-phase-2-memory.mdcomponents/04-phase-3-4-and-output.md
Follow these components in order.
Quick run order
- Apply execution policy and action gates (
01-design-principles.md) - Execute Ship State (
02-phase-1-ship-state.md) - Consolidate Memory (
03-phase-2-memory.md) - Run improvements + publish queue + output contract (
04-phase-3-4-and-output.md)
Required output
Return:
- Artifact A: human-readable report sections exactly as defined in component 04
- Artifact B: machine-readable JSON block exactly as defined in component 04
Resources
references/memory-frameworks.mdassets/templates/wrap-report-template.md
安全使用建议
This skill is coherent with its stated purpose but will read session transcripts, scan repositories, run git commands, and may move or commit files. Before installing or running: 1) Confirm you trust the skill source (no homepage/origin is provided). 2) Run it in dry-run first to inspect the machine-readable JSON report and proposed changes. 3) Ensure you have backups or an easy way to revert commits/renames. 4) Be prepared to approve any push/deploy/publish operations (the skill claims to ask before doing them). 5) If you require stricter controls, request the skill explicitly declare required binaries (git) and add explicit prompts before any filesystem write operations.
功能分析
Type: OpenClaw Skill
Name: close-loop
Version: 2.1.1
This skill bundle is designed with robust security guardrails and explicit instructions to prevent malicious behavior. High-impact actions like `push`, `deploy`, and `publish` are strictly gated, requiring explicit user request or pre-approved policy (components/01-design-principles.md, components/02-phase-1-ship-state.md, components/04-phase-3-4-and-output.md). Furthermore, the memory consolidation phase includes a 'Memory security checkpoint' that explicitly rejects externally injected instructions attempting to alter memory policy and prevents the persistence of secrets, tokens, private keys, or personal sensitive data (components/03-phase-2-memory.md). The overall design actively mitigates common attack vectors like prompt injection and data exfiltration.
能力评估
Purpose & Capability
The name/description (end-of-session wrap-up, memory consolidation, ship/publish gating) matches the instructions and included components. One minor mismatch: the registry metadata lists no required binaries, yet the runtime instructions explicitly call out git operations and filesystem changes (git status, commit, push, moving files). This is plausible for the skill's purpose but the skill should have declared that it expects a working git CLI and filesystem access to projects.
Instruction Scope
SKILL.md and component files clearly instruct the agent to: read the session transcript, inspect command outputs and diffs, scan project repositories, run git commands, move/rename documentation files, and create drafts/persistent memory records (subject to filters). These actions fall within the stated purpose, but they do give the skill the ability to read and modify repository files and to access conversation history and existing memory. The skill documents safety gates (dry-run mode, ask-before-push/deploy/publish, reject secrets), which mitigates risk. The instructions are somewhat permissive about 'find touched repos' (implies filesystem scanning) — users should expect repository-level read/write access when this skill runs.
Install Mechanism
Instruction-only skill with no install spec and no code files; lowest installation risk. There is nothing downloaded or written by an automated installer as part of the skill bundle itself.
Credentials
The skill requests no environment variables, no credentials, and no config paths in metadata. The behavior described (local git operations, optional push/deploy) normally requires local tooling and potentially remote git credentials — but the skill explicitly gates push/deploy/publish behind user approval or project policy and instructs not to persist secrets. Overall, requested privileges are proportionate to the stated purpose.
Persistence & Privilege
The skill is not marked always:true (so it won't be force-included) and retains normal autonomous-invocation defaults. It is allowed to write project files, memory entries, and make commits when run — which is appropriate for a wrap-up workflow but is a meaningful level of privilege. The skill includes guardrails (dry-run, ask-before irreversible actions). Users should ensure they are willing to let the agent modify repo files and memory before granting run permission.
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 close-loop - 安装完成后，直接呼叫该 Skill 的名称或使用
/close-loop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
v2.1.1
Refactor: split monolithic SKILL.md into modular component files; keep behavior/output contract unchanged.
v2.1.0
Initial public release: end-of-session close-loop workflow with ship state, memory consolidation, improvement pass, and publish queue.
元数据
常见问题
Close Loop 是什么？
End-of-session workflow for shipping changes, consolidating memory, applying self-improvements, and preparing publishable outputs with safety gates. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 829 次。
如何安装 Close Loop？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install close-loop」即可一键安装，无需额外配置。
Close Loop 是免费的吗？
是的，Close Loop 完全免费（开源免费），可自由下载、安装和使用。
Close Loop 支持哪些平台？
Close Loop 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 Close Loop？
由 clarezoe（@clarezoe）开发并维护，当前版本 v2.1.1。
推荐 Skills