← 返回 Skills 市场
141
总下载
0
收藏
0
当前安装
3
版本数
在 OpenClaw 中安装
/install agentfiles
功能描述
Use this skill when you need to publish, fetch, search, list, share, or watch AgentFiles artifacts from Codex, Claude Code, OpenClaw, or other agent runtimes...
使用说明 (SKILL.md)
AgentFiles
Use this skill for runtime-facing AgentFiles work:
- publish files or text as artifacts
- fetch artifact content or metadata
- search or list recent artifacts
- create share links
- verify the current principal with
whoami - run polling
watchloops for sidecars or wrappers
Do not reimplement AgentFiles API calls inside the skill unless the user explicitly asks for a direct API path. This skill is a thin wrapper around the existing CLI.
Workflow
- Run AgentFiles commands directly:
agentfiles \x3Csubcommand> .... - Pass argv directly. Never build shell strings around AgentFiles commands.
- If a command needs auth or namespace context, start with
agentfiles whoamioragentfiles config --show. - Prefer
agentfiles setupfor default onboarding. Useconnect \x3Cruntime>only when the user wants a dedicated runtime credential. - For
watch, remember that V1 is polling-only. Readreferences/runtime-notes.mdwhen you need caveats or troubleshooting.
Common Patterns
- Verify auth:
agentfiles whoami - Show config:
agentfiles config --show - Publish text:
agentfiles publish --content "..." --title "..." - Publish file:
agentfiles publish ./path/to/file -n \x3Cnamespace> --title \x3Ctitle> - Fetch content:
agentfiles get \x3Cartifact-id> - Fetch metadata:
agentfiles get \x3Cartifact-id> --meta - Search:
agentfiles search "\x3Cquery>" -n \x3Cnamespace> - List:
agentfiles list -n \x3Cnamespace> - Share:
agentfiles share \x3Cartifact-id> - Watch:
agentfiles watch -n \x3Cnamespace> --json
Handoff
- Hand off with content:
agentfiles handoff codex --content "Please review this patch" - Pipe content:
echo "review notes" | agentfiles handoff codex - Thread a conversation:
agentfiles handoff codex --content "..." --thread pr7-review - Reply back:
agentfiles handoff claude_code --reply-to-artifact-id \x3Cid> --content "Looks good" - Hand off a file:
agentfiles handoff codex ./review.md - Search a thread:
agentfiles search "pr7-review" -n \x3Cnamespace>
Some runtimes may expose this as /handoff. Slash syntax is sugar, not a dependency.
Read references/commands.md for the command matrix. Read references/runtime-notes.md for auth, browser-based connect, polling caveats, and sandbox/network notes.
Behavior
- Prefer an installed
agentfilesbinary onPATH. - If it is unavailable, fall back to the published
agentfiles-clipackage through npm. - Expect network approval when npm needs to download the published CLI package.
- Keep
setupas the default onboarding path andconnectas the advanced/manual path. - Credentials should come from the browser-approved CLI flow and the local
~/.attach/config.jsonfile. - Do not ask the user to paste API keys into the skill or inline them in commands unless they explicitly choose the manual env-based path.
- Preserve CLI behavior. Do not reinterpret command output unless the user asks for a reformatted result.
安全使用建议
This skill is a CLI wrapper that will upload or share files and conversation content to the AgentFiles/MCP service. Before installing or using it: (1) Understand that publishing can send verbatim conversation text and files off the machine—do not publish secrets or private data; (2) Expect the installer to fetch agentfiles-cli from npm (network access) and a browser-based onboarding flow that writes credentials to ~/.attach/config.json; (3) Confirm every handoff/publish operation when prompted; (4) If you require stricter control, run agentfiles whoami / agentfiles config --show to verify the active identity and namespace before publishing; (5) If you need greater assurance, review the agentfiles-cli package on npm (and its source) before allowing the install or network access.
功能分析
Type: OpenClaw Skill
Name: agentfiles
Version: 1.0.2
The skill is a legitimate wrapper for the 'agentfiles-cli' utility (agentfiles.io), designed to facilitate sharing artifacts and context between different AI agent runtimes. The instructions in SKILL.md and references/runtime-notes.md emphasize security best practices, such as avoiding shell string construction to prevent injection and requiring explicit user confirmation before publishing data. No evidence of malicious intent, unauthorized data exfiltration, or hidden execution was found.
能力评估
Purpose & Capability
Name/description match the requested resources: node + agentfiles CLI (or npx fallback) and the ~/.attach/config.json CLI config are appropriate for a CLI wrapper that publishes/fetches artifacts across runtimes.
Instruction Scope
Runtime instructions consistently direct the agent to use the agentfiles CLI or the MCP artifact_publish tool for handoffs. The skill explicitly allows publishing conversation content and files (including verbatim content) to the AgentFiles/MCP server — this is expected for a handoff/publish skill but is a privacy consideration. The skill requires user confirmation before publishing per its rules.
Install Mechanism
Install uses the npm package agentfiles-cli (creates an agentfiles binary). This is an expected, reasonable mechanism for a Node-based CLI. It will require network access on first install; no obscure download URLs or archives are used.
Credentials
No environment variables or external credentials are requested, but the skill reuses local CLI config (~/.attach/config.json) which will contain tokens/credentials obtained via a browser-approved flow. Access to that config and the CLI's auth is proportionate to the skill's purpose but could expose existing credentials/artifact visibility if misused.
Persistence & Privilege
The skill is not always-enabled and does not request elevated platform privileges. It does not modify other skills' configuration. Autonomous invocation is allowed by default but is typical for skills and not, by itself, a flag.
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 agentfiles - 安装完成后，直接呼叫该 Skill 的名称或使用
/agentfiles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
v1.0.2
- No user-facing changes in this release.
- Internal version bump only; no updates to documentation or code.
v1.0.1
- Removed all helper and wrapper scripts for the AgentFiles CLI.
- Skill now runs AgentFiles commands directly via the installed CLI, simplifying command usage.
- Updated documentation to reflect the direct use of agentfiles CLI instead of wrapper scripts.
- Preserved core behaviors and usage patterns, but dropped all internal script-based command resolution and onboarding.
v1.0.0
- Initial release of the agentfiles skill, providing a wrapper around the AgentFiles CLI.
- Enables publishing, fetching, searching, listing, sharing, and watching AgentFiles artifacts across multiple agent runtimes.
- Prefers using a locally installed agentfiles binary, falling back to the agentfiles-cli npm package when necessary.
- Includes helper scripts for running commands and setting up onboarding or credential flows.
- Preserves CLI behaviors without reimplementing API calls, unless specifically requested.
元数据
常见问题
AgentFiles 是什么？
Use this skill when you need to publish, fetch, search, list, share, or watch AgentFiles artifacts from Codex, Claude Code, OpenClaw, or other agent runtimes... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 141 次。
如何安装 AgentFiles？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install agentfiles」即可一键安装，无需额外配置。
AgentFiles 是免费的吗？
是的，AgentFiles 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
AgentFiles 支持哪些平台？
AgentFiles 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 AgentFiles？
由 hammadtq（@hammadtq）开发并维护，当前版本 v1.0.2。
推荐 Skills