[BUG] Specify model for insights command

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

My organization uses AWS bedrock to manage our Claude Code and I don't have access to Claude Opus models. When I use /insights the end report is broken because buttons like "Features to Try" and "Where Things Go Wrong" don't work because there is no output from Opus for the buttons to link to.

What Should Happen?

Claude should be able to detect when a model is unavailable (because of an AWS bedrock authorization error or otherwise) and attempt to complete the request with an alternative model such as Sonnet. Or I should be able to specify what model I would prefer the insights be compiled with.

Error Messages/Logs

## Snippet from my exchange with Claude Code:

The report has two problems:

  1. Empty content — The /insights skill tried to use the Opus model to analyze your sessions, but hit the same AWS authorization error you mentioned. All content sections (Features to Try, What You
  Work On, etc.) are empty because the AI analysis failed.
  2. Missing anchor IDs — The nav buttons link to #section-features, #section-work, etc., but none of those id attributes exist in the HTML, so clicking them does nothing.

  The root cause is the Opus model access issue. Let me check the insights skill to see if it can be configured to use Sonnet instead.

  Searched for 8 patterns, read 4 files (ctrl+o to expand)

⏺ The /insights skill is built into the Claude Code CLI itself, not something I can easily edit. The problem is clear though: the skill hardcoded Opus as the analysis model, which your AWS Bedrock
  role doesn't have access to, so all the AI-generated sections came back empty.

  Your options:

  1. Get Opus access added — ask whoever manages your AWS Bedrock permissions to add anthropic.claude-opus-4-6-v1 to your role's allowed models.
  2. Re-run with a workaround — the /insights skill accepts a model argument. Try running:
  /insights Use claude-sonnet-4-6 instead of Opus for all analysis.
  2. You already passed that as an arg in this session — the issue is the skill's internal hardcoding, not the prompt.
  3. File a bug — the /insights skill should gracefully fall back to Sonnet when Opus isn't available. You can report it at https://github.com/anthropics/claude-code/issues.

  The charts that do show data (tool usage, languages, time of day, errors) are rendered from raw session data and work fine — only the AI-analyzed sections (Features to Try, Friction, Wins, etc.) are
   blank.

Steps to Reproduce

/insights when using an AWS bedrock environment with Claude Opus disabled

Notably an alternative prompt like below did not prevent the issue from occurring.

``/insights Don't use the Opus model as I don't have permissions to use that model. Use Sonnet 4.6 instead.``

Claude Model

Sonnet (default)

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

2.1.87 (Claude Code)

Platform

AWS Bedrock

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

Workaround:

I prompted Claude with:

``❯ Could you retry the insights process with multiple sonnet subagents?``

and that gave me the desired results. Notably I attempted to prevent this issue by prompting for the sonnet model up front (as noted in the steps for recreation), this workaround only succeeded after calling out issues in the report and forcing Claude to realize Opus was unavailable.