Prompt Genie for Claude Code

Stop paying to re-read the same code. Prompt Genie remembers what Claude Code has already read this session, so identical re-reads cost ~15 tokens instead of the whole file.

npx prompt-genie

That's it. Enter your email, restart your editor, and context memory runs silently on every session.

The problem

Claude Code re-reads files and re-runs the same read-only commands constantly. git status between every step, the same source file on turn 2, turn 6, turn 11. Every repeat is sent to the model again, and you pay for all of it. On a long session in a real codebase that's tens of thousands of duplicate tokens.

What Prompt Genie does

It installs a set of Claude Code hooks that fingerprint every file read and read-only command output per session. When Claude reads something it has already seen this session and the content is byte-for-byte unchanged, the hook replaces the output with a short note:

✓ Already in context, unchanged since your last read. 420 lines.

Claude still knows the file is there (it's earlier in the conversation), but you don't pay to send it again.

Why it's safe

The check runs after the tool executes, comparing a hash of the fresh output against what was already delivered this session. If anything changed, a file was edited or a command's output differs, it passes straight through untouched. There are no stale caches and no TTLs to tune.

• Per-session. State is keyed by session id, so concurrent Claude Code windows never interfere with each other.
• Compaction-aware. After /compact or /clear, the memory resets because the files are no longer in context.
• Range-aware. Partial reads of large files are tracked independently of full reads.
• Local-only. Fingerprints (hashes) are stored on your machine. File contents are never stored or transmitted.

Installing

npx prompt-genie

The installer:

1. Copies the hooks to ~/.prompt-genie/hooks
2. Wires them into ~/.claude/settings.json (PreToolUse, PostToolUse, UserPromptSubmit, SessionStart, Stop)
3. Verifies your plan

Restart Claude Code (or your IDE) after installing so the new hooks load.

Seeing your savings

At the end of each session, savings sync to your dashboard at prompt-genie.com: tokens saved, hit rate, and a per-codebase breakdown over time.

For a quick local view any time:

node ~/.prompt-genie/hooks/stats.cjs

Plans

Context memory is a paid feature. On the free plan Prompt Genie runs in measure-only mode and shows you, at the end of each session, how many tokens you would have saved. Activate at prompt-genie.com/pricing.

Uninstalling

Remove the Prompt Genie entries from ~/.claude/settings.json and delete ~/.prompt-genie/.

How your data is handled

The hooks store only content hashes and aggregate counters locally under ~/.claude/. At session end, aggregate stats (tokens saved, hit/miss counts, and the project folder name) are sent to your dashboard. File contents and command outputs never leave your machine.

MIT © Prompt Genie