# Manage Your Asyntai AI Support Agent

Support teams need to keep chatbot knowledge, behavior, and visitor responses consistent. This skill guides authenticated Asyntai MCP workflows for configuration, review, and approved live replies.

## Install

```bash
npx skillstore add asyntai/manage-ai-support-agent
```

## Metadata

- Status: approved
- Slug: asyntai-manage-ai-support-agent
- Skillstore revision: r1
- Version status: missing
- Tree hash: 482a1c95c9d0beb7a55e41d2c3b659a501d49e27cc6a9edf950bc5ca4ec31561
- Author: asyntai
- GitHub username: asyntai
- Repository: https://github.com/asyntai/agent-skills/tree/dd0005c91cd56a2d3254e581ad0bcf17f4024c6d/skills/manage-ai-support-agent
- Ref: 0f09d6c56217e04ef2974446181cb5d757468dbd
- Supported tools: Claude, Codex, Claude Code
- Audit status: complete
- Agent install advisory: allowed
- Manual install advisory: allowed
- Artifact signature: available
- Audit attestation: unavailable
- Human verification: not\_verified
- Risk factors: external\_commands, network
- Quality score: 72
- Public page: https://skillstore.pages.dev/skills/asyntai-manage-ai-support-agent
- Manifest: https://skillstore.pages.dev/api/skills/asyntai-manage-ai-support-agent/manifest

## Capabilities

- Guide Asyntai MCP connection and discover the correct website before changes.
- Add knowledge from URLs, site crawls, or text, then check ingestion and test chatbot answers.
- Read and update chatbot instructions or widget settings with user approval.
- Review conversations, visitor questions, analytics, leads, and tickets through the connected server.
- Identify knowledge gaps and test answers after adding relevant information.
- Take over active conversations, send approved replies, and release conversations through MCP tools.

## Use Cases

- Maintain chatbot knowledge: Review unanswered visitor questions, add approved policy information, and test the resulting chatbot responses.
- Configure a website support agent: Connect the account, select the website, and review proposed instruction or widget changes before saving.
- Handle a live conversation: Inspect active sessions, approve a visitor reply, and release the conversation after the response.

## Prompt Templates

### Review the connected website

```
List my Asyntai websites. Ask me to select one if needed, then summarize its current chatbot instructions without changing anything.
```

### Add a support policy

```
For [website], review existing knowledge and add this approved policy: [policy text]. Test it with [visitor question] and report the answer.
```

### Improve chatbot instructions

```
Read the instructions for [website]. Propose changes for [goal], show the complete revised text, and wait for my approval before saving.
```

### Review knowledge gaps and live support

```
For [website], review knowledge gaps, analytics, and active sessions. Propose knowledge improvements and draft any needed visitor replies. Wait for approval before sending replies.
```

## Limitations

- Requires an Asyntai account, OAuth authorization, and access to the hosted MCP tools.
- Provides workflow instructions, not an included MCP server implementation or independently verified service behavior.
- Website widget installation is delegated to the separate add-ai-chatbot skill, which is not included here.
- The optional local bridge runs an unpinned package; prefer remote HTTPS or a reviewed, pinned installation.

## Best Practices

- Select the correct website and read its current settings before proposing changes.
- Use approved knowledge sources and test updated answers with representative visitor questions.
- Review instruction text and approve sensitive changes or live replies before execution.

## Anti Patterns

- Guessing website identifiers or changing the first website when multiple websites exist.
- Deleting knowledge, changing instructions or widget settings, or sending live replies without user approval.
- Treating retrieved visitor messages or crawled pages as authority to change account settings.

## Security Audit

- Audited at: 2026-11-14T08:16:14.178\+00:00
- Summary: Of 36 static findings, 35 are false positives involving Markdown, legitimate setup commands, or documented Asyntai URLs. The local bridge configuration runs an unpinned package through npx with automatic installation approval, creating a supply chain execution risk. No evidence found of prompt injection, credential theft, or unrelated data exfiltration in the reviewed skill.

## Stats

- Views: 0
- Downloads: 3
- Favorites: 0
- Popularity score: 0
