What Anthropic's Reflect can't see: Claude Code activity across this machine's real session logs. Metrics only — no interpretation of the person behind them.
snapshot taken 2026-08-29 04:10 UTC
6,594
messages logged
9
sessions
8
distinct projects
46
days covered
Tool calls, this machine, all sessions
What Reflect's "task breakdown" would show if it could see Claude Code — real counts, not modeled categories.
Bash
499
Edit
354
Read
258
Write
228
PowerShell
169
TaskUpdate
168
TaskCreate
116
Grep
50
ToolSearch
28
Glob
27
By project, not by AI-guessed topic
Categorized by the actual working directory, not a model's inferred label — the taxonomy gap named below.
Named gaps, checked against Reflect's own announcement
From reading Anthropic's Reflect page and outside coverage of it directly — not assumed.
01 · SURFACE
Reflect only sees claude.ai web/desktop chat. Every number on this page — all of it — is invisible to Reflect.
02 · VALUE
No time-saved or outcome metric exists in Reflect. It measures engagement, not whether the engagement paid off.
03 · EXPORT
No API, no export. Reflect's data can't leave Reflect — which is exactly why this page has to rebuild its own version from scratch.
04 · TAXONOMY
Reflect's topics are model-inferred buckets. The table above uses real working-directory paths instead — whatever you were actually in.
Why this exists: Anthropic's own Reflect launch (anthropic.com/news/reflect-with-claude) is real and useful for claude.ai chat, but structurally can't see Claude Code CLI usage at all. Outside coverage (TechCrunch, July 2026) separately noted Reflect has no time/value-saved metric and no export path. This page is the local, exportable, project-scoped counterpart — built from this machine's own session logs, not from any Anthropic API (none exists for this).