Prism

Session intelligence for Claude Code. Find out why your sessions are burning tokens, and fix them.

The problem

Your Claude Code sessions are silently wasting tokens and you don't know why.

Running PRISM against real session data from a single machine found:

- A project with 6738% CLAUDE.md re-read cost in one session: a 237-line file being re-read on every tool call
- A project where CLAUDE.md re-reads consumed 480% of total session tokens: more tokens on instructions than on actual work
- 4 migration file edits in a project that had a rule saying never to touch them. The rule existed, Claude ignored it
- 5 consecutive tool failures in a single session with no diagnosis

None of this was visible before PRISM. The token counter just said you hit your limit.

What PRISM does

PRISM reads Claude Code's session files from ~/.claude/projects/ (the same files Claude Code writes automatically) and tells you three things:

1. Why your tokens are disappearing: CLAUDE.md re-read costs, retry loops, compaction losses, sidechain waste
2. Whether your CLAUDE.md rules are actually being followed, or silently ignored mid-session
3. Exactly what to change: concrete diff recommendations, not generic advice

You keep using Claude Code exactly as normal. PRISM is the tool you run after.

Quick start

Then just ask Claude: "analyze my Claude Code sessions"
Note: PRISM needs to be installed via pip to work:
pip install prism-cc
If pip isn't installed, Claude Code will detect this
and walk you through the installation automatically
before running the analysis.

What you'll see

Followed by the advisor:

📊 HTML Dashboard Preview

*The HTML dashboard. Open in any browser with prism dashboard*

*Expand any project to see dimension scores, top issues, and CLAUDE.md…