The replay-triage skill connects a Pendo replay to your code, so you can move from a bug a visitor experienced toward a proposed code fix in a single workflow. How far it can take you depends on which tools your AI client is connected to (see Prerequisites).
Use cases
Use the Replay Triage skill to:
- Diagnose a bug a visitor experienced, starting from the replay that captured it.
- Start from a bug report in your project management tool, such as Jira or Linear, that links to a replay.
- See what went wrong during a frustrating or failed session.
- Move from a reported problem to a reviewed code fix faster.
How it works
Given a replay URL, or a ticket that links to one, the skill follows the same sequence each time:
- Resolve the source. It accepts a replay link directly, or pulls one from a linked ticket.
- Gather devlog events. It collects the console errors and failed network requests captured during the session.
- Triage the errors. It groups related errors into a short list of distinct issues, ranked by severity, rather than returning every log line.
- Map errors to your code. It locates the relevant files in your current project for the top issues.
- Propose a fix. It presents a change with a short diff for your review, before editing anything.
- Prompt for a pull request. After you approve and the change is applied, it summarizes the result and prompts you to open the pull request.
Note: The extent of the skill's capabilities depends on your AI client's connections. Mapping errors to code and proposing a fix require a connected code host, and starting from a ticket requires a connected ticketing tool, such as Jira or Linear. Without those, the skill still retrieves and triages the session's errors, but it can't locate the relevant code or propose a fix. Regardless of your connections, the skill never applies edits or opens a pull request without your confirmation.
Prerequisites
To run the Replay Triage skill, you need:
- Session Replay turned on for the app you're triaging replays for, with console or network log capture enabled in app settings. The skill reads these logs, so at least one must be on.
- A Pendo connection through MCP. The skill reads replay devlog data from Pendo, so an active Pendo MCP connection is required. The skill won't run without it.
- Access to your project's code in the working directory where you run the skill, so it can match errors to files and propose a fix.
- An AI client that supports skills, such as Claude or Cursor.
The following connections are optional and unlock more of the workflow:
- A ticketing tool connected in your AI environment. If your AI client is connected to a ticketing tool (such as Jira through the Atlassian MCP or Linear through the Linear MCP), you can start from a ticket instead of a replay link. The skill pulls the replay link and bug context from the ticket.
- A code host connected in your AI environment. Connect your code host (such as GitHub) so the skill can cross-reference the session's errors against your codebase.
Install the skill
Download the replay-triage skill file and add it to your preferred AI coding assistant. For installation steps, see the documentation for your tool. For example:
Use the Replay Triage skill
After you download the skill and provide it to your AI client:
- Provide a replay link or a ticket. Supply a replay player URL, a saved clip URL, or a ticket reference (a Jira key or URL, or a GitHub issue URL) that contains a replay link.
- Add a focus hint (optional). Include a short description of the problem, such as "login fails" or "500 on save," to point the triage at the right errors. When you start from a ticket, the skill uses the ticket description for this context.
- Run the skill. The skill parses the replay, retrieves the devlog events, and groups them into a ranked short list of issues. Retrieving logs can take a moment.
- Review the issues and proposed fix. The skill presents the top issues with the supporting evidence and, where it can locate the code, a proposed fix with a short diff. Review the proposal before continuing.
- Approve the change. If the fix looks right, approve it and the skill applies the edit to your local files. If it needs adjustment, ask for changes and review again.
- Verify and open a pull request. After the change is applied, the skill runs your project's existing type checks or tests when they're configured, then summarizes what changed. Open the pull request yourself when you're ready.
Tip: A focus hint makes triage faster and more accurate on noisy sessions. If you know the symptom, name —for example "UI freezes after save."
What the skill doesn't do
The skill is scoped to keep you in control of changes to your code and tickets:
- It doesn't open pull requests. It prepares the change and prompts you to open the pull request when you're ready.
- It doesn't modify tickets. It reads a linked ticket for context, but doesn't comment on, transition, or change it.
- It doesn't make broad changes. Fixes are scoped to the bug in the replay, with no unrelated refactors or cleanup.
- It doesn't guess at file locations. If a stack trace is minified or an error is generic, it tells you it can't map the error rather than choosing a file.
- It needs a replay URL to run. A bug description on its own isn't enough to start triage.
Troubleshooting
No errors found in the session. The skill widens its search to include lower log levels and all network traffic. If a session still has no captured errors or failures in the recorded window, the skill tells you, and you can try a different replay or widen the time window.
The replay link is missing its time range. Copy the player link again from Session Replay. The player URL exported from Pendo includes the start and end time that the skill needs.
A ticket links to more than one replay. The skill lists the candidate replays with context and asks which one to triage.
The error can't be matched to your code. If a stack trace is minified or the error message is generic, the skill says it can't locate the source instead of guessing. Add a focus hint about which part of the app is involved, and run it again.