- Home
- Skills
- Agents & Orchestration
- CodeMap Pro: Claude Code reads 65% less
More screenshots
Works with the AI Agents you already use
CodeMap Pro: Claude Code reads 65% less
A code map for Claude Code. Claude queries a graph of your project first, then reads only the lines it needs instead of whole files. Auto-refresh hook and a live, local token analytics dashboard included.
$6
CodeMap Pro: Claude Code reads 65% less
Example session with this skill installed
Where does this project measure the token savings from the Claude Code logs?
- Read your context and instructions
- Compiled the codemap pro
Summary
The savings are measured in live_measure() in scripts/measure.py. I found it with one map query and read only the lines around it, instead of opening the whole files.
Where it happens
| What | File | Line |
|---|---|---|
| Reads the session logs, counts map queries and the targeted reads after them | scripts/measure.py | 793 |
Recognises a map query (graphify query, explain, path) and checks that it answered | scripts/measure.py | 181 |
Finds partial reads in the shell (sed -n, head, tail, grep with context) | scripts/measure.py | 432 |
Works out which project a query was about, from its graph.json path | scripts/measure.py | 201 |
| Counts each file whole once per session, at most 2,000 lines | scripts/measure.py | 484 |
| Runs the measurement every 3 seconds while the dashboard is open | scripts/dashboard.py | 35 |
How it works
live_measure() walks every *.jsonl session log in ~/.claude/projects. A map query counts when graphify actually answered. For the next 12 tool calls it looks for partial reads of the files that answer named: Read with offset or limit, or sed -n, head, tail and grep with context in the shell. Each file is compared with reading it whole once per session, at most 2,000 lines (the Read tool's limit), so a second excerpt of the same file does not count the whole file again. A file the session read whole anyway saves nothing, and a read that printed the whole file is not counted. Results are kept in total, per day and per project, and every tool call counts once by its id. Each log is read line by line, a log that did not change since the last run is not read again, and a log that keeps growing is followed from where the last read ended.
What this answer cost
Measured with the built-in question test on this repository: a keyword search reading the 5 best matching files whole would have cost
18,874 tokens. The map query plus targeted excerpts cost
3,009 tokens, which is
84% fewer.
Next steps
- Open the dashboard with
dashboard.pyto see the same numbers for your own sessions. - Run
codemap.py --folder <project> --measure "<question>"to test any question yourself.
Connects securely to your tools. The creator never sees your data.
About this skill
Claude Code spends most of its context reading files. Ask where something happens in a big project and it greps, opens five whole files, and reads thousands of lines it never needed.
CodeMap Pro gives Claude a map instead. It builds a graph of your project (functions, classes, imports, calls, docs, media and the folder tree) with graphify, and teaches Claude to query the map first and then open only the few lines it points to.
In our own Claude Code work it cut file reading by 65.6%: 70,473 tokens of whole files became 24,259 tokens of map queries and targeted excerpts, over 25 map queries and 9 targeted reads. That is a small sample, measured on 2026-11-10. Your number will differ by project, and the built-in dashboard measures it on your own sessions.
What you get
- The skill: Claude checks the map before it opens files, then reads only the lines that matter.
- Auto-refresh hook: when a session starts in a project, the map refreshes in the background in a few seconds and Claude is told where it is.
- Live token analytics dashboard: it reads your Claude Code logs every 3 seconds while it is open, so a new map query shows up within seconds. A live scanner shows the tool calls of every session, a feed lists the latest map queries and targeted reads, and the figures roll to their new values as they change. Real savings per day and per project, a token bridge from whole files to what Claude read, the API cost avoided at the list price of the model you use (or your own rate), question tests, map health and an interactive graph viewer.
- One-command installer for macOS and Linux, with a clean uninstall.
- Workspace mode: map a folder of many projects at once, plus one merged graph.
Privacy and cost
No LLM calls, no API keys, no cost per run, no file-count limit. Your code never leaves your machine, and nothing is written into your project folders. The only download is graphify itself (uv tool install graphifyy), and only if it is missing.
Requirements
Claude Code, uv, macOS or Linux.
Does not index the internal content of images, videos, or PDFs. Project maps are stored in ~/.codemap/ or workspace subfolders, not the project root.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 1 day ago
Requires
Permissions
Terminal: runs the map builder and graphify queries. Read files: reads the project to build the map and Claude Code's session logs in ~/.claude/projects for the savings dashboard. Write files: maps go to ~/.codemap or <workspace>/graphify-out, and the optional installer adds one SessionStart hook to ~/.claude/settings.json (a backup is saved). Network: only the installer, to run 'uv tool install graphifyy' when graphify is missing; the dashboard listens on 127.0.0.1 only. Environment variables: optional CODEMAP_WORKSPACE and CODEMAP_HOME settings.