// lean ai ops

Your AI is running fat.

We built the cure — because we got the bill first.

zsh — claude-diet
Run the audit →Read the receipts

Big Tech wants you to buy more context. We want you to need less.

// how it works

01 / run it

One command. No install. No account.

Paste it in your terminal. The script fetches, runs, and tells you exactly what your Claude Code setup loads before you type a prompt.

02 / diet card

Your runtime diet card prints immediately

No dashboard. No account. The card prints right in the terminal — largest surfaces, hidden memory, hooks, skills, and the top suspect identified.

03 / rules + report

Rules generated. Report saved. Done.

Three next-session rules written from your actual setup. A full report folder dropped in your project directory — card, worksheet, baseline, passport.

diet — audit
$ bash claude-runtime-diet-report.sh

Claude Code Runtime Diet
Auditing context surfaces... ◐

[1/4] Baseline generated.
[2/4] Extracting top surfaces and
      hidden memory...
[3/4] Writing worksheet, card,
      passport, rule, index...
diet — card
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓┃                   My Claude Runtime Diet Card                          ┃┣━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┫┃ Largest visible surface                                                ┃┃ ~/.claude/projects - 819MB                                             ┃┃                                                                        ┃┃ Hidden project memory total     biggest invisible cost                 ┃┃ 818.2MiB across 24 projects                                            ┃┃                                                                        ┃┃ Largest hidden project memory                                          ┃┃ my-big-project - 256MB (381 files)                                     ┃┃                                                                        ┃┃ Hooks: 2 | MCP servers: 2 | Skills: 35 | Agents: 2                     ┃┃                                                                        ┃┃ Top suspect                                                            ┃┃ Long-lived session transcript; old branches make the next session      ┃┃ heavier than the prompt looks.                                         ┃┗━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┛
diet — lean
[4/4] Done.

Runtime Diet Report written to:
  ~/projects/my-app/
    runtime-diet-report-20260508

Open first:
  my-runtime-diet-card.md
  NEXT-SESSION-RULE.md

Next session rules (generated):
1) Start fresh, don't drag old branches
2) Prune ~/.claude/projects (818MB)
3) Cap MCP + skills + agents per task

Layer 1

Runtime Diet Autopilot

Audit your Claude Code context in 60 seconds. Free.

One shell script scans your CLAUDE.md files, hooks, and memory surfaces, then hands you a Runtime Diet Card — exactly where your tokens are going and what to cut first. No account. No email. No catch.

Get the script →
clone + run
$ git clone https://github.com/skinny-cloud/skinny.cloud$ cd skinny.cloud$ bash tools/runtime-diet-report/claude-runtime-diet-report.sh
0K tokens
loaded pre-prompt
0 projects
in production
0s audit
no signup

One resumed session recreated 389,169 tokens of cached context before it answered a single prompt. The audit makes that visible.