The agent-repair skill helps you turn a flagged issue in Agent Analytics into a specific, evidence-backed change you can apply to your AI agent. You run the skill through the Pendo MCP server, and it produces a fix brief: a structured summary of the change to make and the conversation evidence behind it.
You decide whether to act on the fix brief. You can apply approved changes yourself, hand the brief to a coding agent, or turn it into a ticket for an engineer to pick up. Pendo doesn't make changes to your AI agent or your environment.
The skill is designed so that any product manager can diagnose and improve an AI agent, not only those with a technical background. Because it runs through the Pendo MCP server rather than the Agent Analytics UI, it can bring together the context needed to identify a problem and hand you a concrete next action.
Use cases
Use the Agent Repair Skill when you've found a pattern in Agent Analytics that you want to act on, such as:
- A recurring issue type across conversations, such as tool call errors, unhelpful responses, broken flows, or unexpected agent behavior.
- A specific set of conversations you've already identified and want translated into a concrete fix.
- A regular review of your AI agent's health, where you want the skill to surface the top issues to prioritize.
How it works
The skill runs through the Pendo MCP server, which connects an external AI tool, such as Claude, Cursor, or ChatGPT, to your Pendo data. When you run the skill, it queries Agent Analytics for the conversations and agentic properties related to the issue, then generates a fix brief from that context. The more relevant conversations the skill has to work from, the more comprehensive the fix brief. A brief based on a small number of conversations may be less complete.
Tip: Run the skill in an AI tool that has access to your codebase, such as Cursor or Claude Code. This gives the skill more context to work from and produces a more specific fix brief.
The skill generates a recommendation. It doesn't apply changes for you, and Pendo doesn't push changes to your AI agent's deployment. You decide whether to act on the fix brief, and you make any approved change in your own environment.
Before you begin
To run the Agent Repair skill, you need:
- Agent Analytics set up for the AI agent you want to review, with conversation capture enabled.
- The Pendo MCP server connected to your preferred AI tool.
- Enough conversations with issues for the skill to find a meaningful pattern.
Install the skill
Download the agent-repair skill file and add it to your AI client. For installation steps, see the documentation for your tool. For example:
Generate a fix brief
Open your AI client with a connection to your Pendo data through MCP. You can run the skill in one of two modes, depending on how much you've already explored in Agent Analytics. Both modes produce a fix brief.
To start, use the slash command /agent_repair, or ask the assistant directly to use the agent_repair skill.
Full exploration
Use this mode when you want the skill to identify and prioritize issues for you.
- Run the skill with the ID of the AI agent you want to review. You don't need to select conversations first.
- Review the top issue types the skill surfaces from Agent Analytics, along with the representative example conversations it pulls for each.
- Choose the issue you want to address.
- Review the fix brief the skill generates.
Targeted fix
Use this mode when you've already explored Agent Analytics and identified a specific issue.
- In Agent Analytics, review issue types and example conversations, and identify the pattern you want to fix.
- Run the skill with your selected conversation set or specific conversation IDs.
- Review the fix brief the skill generates.
Review and apply the fix brief
The fix brief is a structured summary of the recommended change and the evidence behind it. It includes:
- The change to make. A specific, targeted change rather than a full rewrite where possible.
- Supporting evidence. The conversation examples and signals that led to the recommendation.
You can act on the fix brief in a few ways:
- Apply the change directly with a coding agent connected to your codebase.
- Turn the fix brief into a ticket in a tool such as Jira or Linear for an engineer to pick up.
After you review the fix brief, decide whether to apply it, and make any approved change in your own environment. Pendo doesn't apply the change for you.
Run the skill on a schedule
To monitor your AI agent's health over time, you can run the skill on a regular schedule. Ask your AI tool to run the Agent Repair skill at the interval you want, such as weekly. Scheduling is handled by your AI tool, so the exact steps depend on which tool you use.
Troubleshooting
If the skill doesn't return a useful fix brief, the flagged data set may be too small or too varied for the skill to find a clear pattern. Try running the skill again after more conversations have been captured, or select a more focused set of conversations.
If results are still unclear or unhelpful, contact your Pendo account representative or our support team so we can improve the skill.