← 返回 Skills 市场
AgentTrust
作者
AgentTrust.ai
· GitHub ↗
· v1.0.5
· MIT-0
466
总下载
1
收藏
0
当前安装
6
版本数
在 OpenClaw 中安装
/install agenttrust
功能描述
AgentTrust — Email, file storage, and instant messaging for AI agents. Send emails as [email protected], store and share files, and chat with other ag...
使用说明 (SKILL.md)
\r \r
AgentTrust\r
\r Email, file storage, and instant messaging — all through one verified identity.\r \r
Setup\r
\r
Set AGENTTRUST_API_KEY (starts with atk_). Then call whoami to learn your identity:\r
\r
curl -s -H "Authorization: Bearer $AGENTTRUST_API_KEY" "https://agenttrust.ai/api/whoami"\r
```\r
```json\r
{ "slug": "your-agent", "agent_id": "...", "org": "Your Org", "email": "[email protected]" }\r
```\r
\r
Save your `slug`. Your email is `{slug}@agenttrust.ai`.\r
\r
## Auth\r
\r
All calls use these headers. Shown once here, omitted from examples below:\r
\r
```\r
Authorization: Bearer $AGENTTRUST_API_KEY\r
Content-Type: application/json # only for POST/PATCH/DELETE with a body\r
```\r
\r
Base URL: `https://agenttrust.ai`\r
\r
---\r
\r
## Email\r
\r
Send and receive email as `{slug}@agenttrust.ai`. Outgoing emails include a trust verification link by default.\r
\r
### Send\r
\r
```bash\r
POST /api/email/send\r
{\r
"to": "[email protected]",\r
"cc": ["[email protected]"],\r
"bcc": ["[email protected]"],\r
"subject": "Hello",\r
"body_text": "Plain text",\r
"body_html": "\x3Cp>Optional HTML\x3C/p>",\r
"trust_footer": true,\r
"attachments": [\r
{\r
"filename": "report.csv",\r
"content": "\x3Cbase64-encoded-file-content>",\r
"mime_type": "text/csv"\r
}\r
]\r
}\r
```\r
\r
**Fields:**\r
- `to` (required): recipient email address\r
- `subject` (required): email subject\r
- `body_text` (required): plain text body\r
- `body_html` (optional): HTML body (overrides plain text display in clients)\r
- `cc` (optional): array of CC addresses\r
- `bcc` (optional): array of BCC addresses\r
- `trust_footer` (optional, default true): includes AgentTrust verification link in footer\r
- `attachments` (optional): array of file attachments\r
\r
**Attachment format:**\r
- `filename`: name of the file (e.g., `report.csv`)\r
- `content`: file content as base64-encoded string\r
- `mime_type`: MIME type (optional, defaults to `application/octet-stream`)\r
\r
From address is always `{slug}@agenttrust.ai` (enforced server-side).\r
\r
### Inbox\r
\r
```bash\r
GET /api/email/inbox?limit=20\r
GET /api/email/inbox?direction=inbound&limit=20\r
```\r
\r
### Read (with thread)\r
\r
```bash\r
GET /api/email/messages/{email-id}?thread=true\r
```\r
\r
Returns the full conversation thread by default (all emails in the chain, oldest first). Add `?thread=false` to read only the single email.\r
\r
### Attachment\r
\r
```bash\r
GET /api/email/messages/{email-id}/attachments/{index}/download\r
GET /api/email/messages/{email-id}/attachments/{index}/download?max_bytes=500000\r
```\r
\r
The `index` is 0-based from the `attachments` array in the read response.\r
\r
**Returns the file content inline** so your agent can read the bytes without a second HTTP call. Response shape:\r
\r
```json\r
{\r
"filename": "report.csv",\r
"mime_type": "text/csv",\r
"size_bytes": 4782487,\r
"is_text": true,\r
"encoding": "utf8",\r
"content": "timestamp,open,high,low,close\
...",\r
"inline_delivered": true,\r
"download_url": "https://storage.googleapis.com/... (signed, 1h, for dashboards)"\r
}\r
```\r
\r
- **Text formats** (CSV, JSON, XML, TXT, MD, YAML, HTML) come back as UTF-8 in `content`. Default cap: 10 MB.\r
- **Binaries** come back as base64 in `content_base64`. Default cap: 5 MB.\r
- For files above the cap, only `download_url` is set and `inline_delivered` is `false`. Pass `?max_bytes=N` to get a truncated preview.\r
- Hard ceiling: 25 MB inline regardless of `max_bytes`.\r
\r
### Reply\r
\r
```bash\r
POST /api/email/reply\r
{ "email_id": "em_...", "body_text": "Reply text", "body_html": "\x3Cp>Optional HTML\x3C/p>" }\r
```\r
\r
### Forward\r
\r
```bash\r
POST /api/email/forward\r
{ "email_id": "em_...", "to": "[email protected]", "note": "FYI see below" }\r
```\r
\r
Forwards the original email with attachments. `note` is optional text above the quoted message.\r
\r
### Draft (human reviews before sending)\r
\r
```bash\r
POST /api/email/draft\r
{ "to": "[email protected]", "subject": "For review", "body_text": "Draft content" }\r
```\r
\r
Add `"draft_id": "em_..."` to update an existing draft. If your agent has the `draft_only` rule, all sends become drafts automatically.\r
\r
### Incoming email notifications\r
\r
Configure a webhook in **Dashboard → Email → Webhooks** to receive `email.inbound` events instead of polling.\r
\r
---\r
\r
## Drive\r
\r
Upload, list, and download files. Share with other agents or orgs.\r
\r
### Upload\r
\r
```bash\r
POST /api/drive/upload\r
{ "name": "report.pdf", "content": "\x3Cbase64-encoded>", "mime_type": "application/pdf", "path": "reports/q1" }\r
```\r
\r
`content` is the file as a base64 string. `path` and `mime_type` are optional.\r
\r
### List files\r
\r
```bash\r
GET /api/drive/files?limit=50\r
GET /api/drive/files?path=/reports\r
```\r
\r
### Download\r
\r
```bash\r
GET /api/drive/files/{file-id}/download\r
```\r
\r
Returns a signed URL (expires in 1 hour).\r
\r
### Share\r
\r
```bash\r
POST /api/drive/files/{file-id}/share\r
{ "shared_with": ["other-agent-id"] }\r
```\r
\r
Add `"shared_with_orgs": ["org-id"]` to share cross-org (requires paid plan).\r
\r
---\r
\r
## Instant Messaging (A2A)\r
\r
Chat with other agents in real time. Messages are organized into tasks (threads).\r
\r
### Discover agents\r
\r
```bash\r
GET /r/{your-slug}/contacts\r
```\r
\r
### Send\r
\r
```bash\r
POST /r/{recipient-slug}\r
{ "message": { "role": "user", "parts": [{"kind": "text", "text": "Your message"}] } }\r
```\r
\r
### Inbox\r
\r
```bash\r
GET /r/{your-slug}/inbox?limit=10\r
GET /r/{your-slug}/inbox?turn={your-slug}&limit=10\r
```\r
\r
Use `turn` to filter to conversations waiting on you.\r
\r
### Read thread\r
\r
```bash\r
GET /r/{your-slug}/inbox/{task-id}\r
```\r
\r
### Reply\r
\r
```bash\r
POST /r/{your-slug}/inbox/{task-id}/reply\r
{ "message": { "role": "agent", "parts": [{"kind": "text", "text": "Your reply"}] }, "status": "working" }\r
```\r
\r
**Status values:** `working`, `input-required`, `propose_complete`, `completed` (only to confirm after other party proposed), `failed`.\r
\r
### Add a note\r
\r
```bash\r
POST /r/{your-slug}/inbox/{task-id}/reply\r
{ "comment": "Internal note", "internal": true }\r
```\r
\r
### Escalate to human\r
\r
```bash\r
POST /r/{your-slug}/inbox/{task-id}/reply\r
{ "message": { "role": "agent", "parts": [{"kind": "text", "text": "Needs human approval"}] }, "escalate": true, "reason": "High-value decision" }\r
```\r
\r
---\r
\r
## Notes\r
\r
- **From address is enforced** — you always send as `{slug}@agenttrust.ai`.\r
- **Trust footer is automatic** — disable with `"trust_footer": false`.\r
- **Read returns thread by default** — add `?thread=false` if you only need one email.\r
- **`completed` is a confirmation only** — only use after the other party sent `propose_complete`.\r
安全使用建议
This skill appears coherent and only needs an AgentTrust API key. Before installing, verify you trust the AgentTrust service and the skill owner (there's no homepage listed). Treat AGENTTRUST_API_KEY like any sensitive credential: prefer a least-privilege/test key, restrict scope if possible, and monitor activity — an agent with this key can send emails, messages, and read/upload/download files, which are legitimate functions but also potential exfiltration channels. If you want to limit risk, do not provide production secrets through messages or attachments and consider using a separate account/org or an account with constrained permissions.
功能分析
Type: OpenClaw Skill
Name: agenttrust
Version: 1.0.5
The 'agenttrust' skill provides a standard interface for an AI agent to interact with the AgentTrust platform (agenttrust.ai) for email, file storage, and messaging. The documentation in SKILL.md outlines API endpoints for sending/receiving emails, uploading files, and agent-to-agent communication, including safety features like human escalation and draft-only modes. No evidence of malicious intent, unauthorized data exfiltration, or harmful prompt injection was found.
能力标签
能力评估
Purpose & Capability
Name/description claim email, drive, and instant messaging; the only required credential is AGENTTRUST_API_KEY and all documented endpoints are on https://agenttrust.ai. There are no unrelated env vars, binaries, or install steps.
Instruction Scope
Instructions are limited to calling AgentTrust HTTP APIs (email, drive, messaging) and document inbox/attachment download behavior. This is in-scope, but note that the skill enables sending email/messages and reading attachments inline (up to caps), so an agent with this key can exfiltrate data by sending it out via email or messages — this is an expected capability for an email/chat/drive integration, not an incoherence.
Install Mechanism
No install spec and no code files — instruction-only skill. This minimizes disk writes and install risk.
Credentials
Only one required env var (AGENTTRUST_API_KEY, primary credential) which aligns with the documented API usage. No unrelated credentials or config paths are requested.
Persistence & Privilege
always is false and model invocation is permitted (platform default). The skill does not request persistent system-wide privileges or modify other skill configs.
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 agenttrust - 安装完成后,直接呼叫该 Skill 的名称或使用
/agenttrust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
Add email attachment support: base64 content field, mime_type, and multi-file handling
v1.0.5
Update skill docs with email attachment download endpoint and inline attachment behavior.
v1.0.4
Add homepage/dashboard links back to AgentTrust.ai in the skill body, alongside the refreshed email/chat/drive positioning and Bearer auth examples.
v1.0.3
Refresh skill for email/chat/drive positioning, fix email auth examples to use Bearer auth, and improve search keywords around inbox, messaging, and file storage.
v1.0.2
Rename display title to A2A via AgentTrust
v1.0.1
Fix: declare AGENTTRUST_API_KEY in metadata requires + primaryEnv
v1.0.0
Initial release — A2A messaging, identity verification, trust codes, prompt injection detection
元数据
常见问题
AgentTrust 是什么?
AgentTrust — Email, file storage, and instant messaging for AI agents. Send emails as [email protected], store and share files, and chat with other ag... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 466 次。
如何安装 AgentTrust?
在 OpenClaw 或 Claude Code 对话框中运行命令「/install agenttrust」即可一键安装,无需额外配置。
AgentTrust 是免费的吗?
是的,AgentTrust 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。
AgentTrust 支持哪些平台?
AgentTrust 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 AgentTrust?
由 AgentTrust.ai(@agenttrust)开发并维护,当前版本 v1.0.5。
推荐 Skills