[BUG] /insights report makes misleading productivity judgments from incomplete local-only data

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?

The /insights report draws conclusions about a user's productivity based solely on local CLI session data, without disclosing that large portions of Claude usage are invisible to it. Specifically, it cannot see:

• Sessions from claude.ai web (claude.ai/code or chat)
• Claude Code sessions on other machines
• Git commits and PR merges made outside a Claude Code CLI session on this machine
• Commits pushed to GitHub repositories — even those made through Claude Code on another device, which should be retrievable via the GitHub API without requiring cross-machine file access

Despite these blind spots, the report generates statements like:

▎ "x sessions, y messages, zero commits — someone spent hours talking to Claude without shipping a single line"

This treats the absence of locally-detected commits as evidence of zero output. A user could have merged multiple PRs that week through Claude Code on another machine or via the web, and receive the same verdict.

What Should Happen?

Expected behavior:

At minimum, scope the language to what was actually measured: "0 commits detected in these 3 local CLI sessions." Do not draw productivity inferences beyond the data scope.

Beyond that, two improvements worth considering:

1. GitHub integration — since Claude Code already operates on git repositories, the report could optionally query the GitHub API for commits and PRs authored by the user, providing a more complete picture without requiring any cross-machine sync.
2. User-configurable data scope — allow users to opt in to cross-machine usage sync (e.g. via their Anthropic account), or at minimum display a visible disclaimer indicating what the report does and does not cover, so users can make informed decisions about how to interpret the data before acting on it.

Actual behavior: The report makes unqualified productivity claims using metrics it knows are incomplete, with no disclosure of data scope.

Error Messages/Logs

Steps to Reproduce

What's Wrong section is already updated with details.

Claude Model

Sonnet (default)

Is this a regression?

No, this never worked

Last Working Version

_No response_

Claude Code Version

2.1.96 (Claude Code)

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

VS Code integrated terminal

Additional Information

_No response_