← All products
🔍

Claude SDK Audit Kit

On June 15, 2026 Anthropic splits the Claude Agent SDK off your subscription. The Agent SDK, claude -p, Claude Code GitHub Actions, and third-party SDK apps draw from a separate $200 monthly credit on Max 20x. For a real 24/7 agent that credit drains in days. Overflow lands on standard API rates on top of the subscription you already pay. This kit is the audit and migration tooling I built to soften that cliff on my own machine, packaged so you can run it on yours. Hand the kit to Claude Code, say "run the audit", and it walks your repo, cron, and launchd, finds every claude -p call and SDK import, estimates monthly cost on each plan, and writes a migration plan you can act on before the deadline.

✓An add-on for the Claude Code agent you already run. Mitigates the June 15, 2026 Agent SDK billing split.
New May 22, 2026 · Initial release covering the June 15, 2026 Agent SDK billing split. Four audit scripts, four mitigation playbooks, migration spec template, and worked example.

Paid Digital Thoughts subscriber?

Yearly subscribers: all products free ($1111.94+ value)

Monthly subscribers: 1 free product per month — subscribe from $29.99/mo

Claim your free copy →
$9.99One-time purchase. 12 months of updates. Yearly subscribers get this free.

What you get

+Four audit scripts that scan your machine for every claude -p call, every Agent SDK import, and every cron/launchd job touching Claude
+Monthly cost estimator with per-plan breakdown (Pro / Max 5x / Max 20x / Team Premium / Enterprise Premium)
+Four mitigation playbooks: drive interactive Claude, second harness, route narrow calls, audit and trim
+Migration spec template, one page per call you decide to migrate
+Worked example from my own stack, sanitized but honest numbers
+AGENT-INSTRUCTIONS.md so Claude Code drives the audit end to end
+No external network calls, no destructive writes, read-only against your machine except for the report file
+Honest shelf-life notes on each mitigation (Mitigation 1 is a 6-12 month bridge, Anthropic will probably patch the easy version)
+12 months of updates as Anthropic ships follow-up pricing changes

Package includes

  • •START-HERE.html: styled entry page, open first
  • •AGENT-INSTRUCTIONS.md: prompt your Claude Code uses to drive the audit
  • •README.md: pitch and overview
  • •QUICKSTART.md: 10-minute DIY install
  • •audit/run-audit.sh: wrapper that runs the four sub-scripts
  • •audit/scan-claude-p.sh: finds every claude -p invocation in scripts, cron, launchd, shell config
  • •audit/scan-sdk-imports.py: finds Agent SDK imports across Python and TypeScript repos
  • •audit/find-cron-leaks.sh: surfaces cron jobs and launchd plists that touch Claude
  • •audit/estimate-cost.py: ballparks monthly cost on each plan with assumptions you can tune
  • •mitigations/01-drive-interactive-claude.md: the AppleScript + Terminal + clipboard paste pattern
  • •mitigations/02-second-harness.md: Codex as insurance + extension
  • •mitigations/03-route-narrow-calls.md: OpenRouter and local-model routing recipes
  • •mitigations/04-trim-checklist.md: KILL, SLOW, MERGE, TRIM, CHEAPER, INTERACTIVE, HARNESS-SWAP labels
  • •templates/migration-spec.md: one-page migration spec per call
  • •templates/audit-report-template.md: shape of the output your agent writes
  • •examples/example-audit-report.md: sanitized worked example from my own machine
  • •CHANGELOG.md: update history

FAQ

Why $9.99 and not the usual kit price?

This is a cheap add-on, not a full kit. The audit scripts are simple, the playbooks are short, the migration spec is one page. The value is timing. If you wait until July to run an audit, your June bill already hurts.

Does this work with Codex or Cursor?

The audit can scan any directory and find imports for any provider, but the cost model and the four mitigations are Claude-specific. If you only use Codex, this kit is not for you.

How accurate is the cost estimate?

It is a ballpark with documented assumptions. Default is 50k input + 5k output tokens per call, 70% Sonnet 4.6 / 30% Opus 4.7 mix, no prompt cache credit. Tweak the constants in estimate-cost.py to match your measured usage. Real bills will usually be lower because of prompt caching.

Will Anthropic patch Mitigation 1 (driving interactive Claude)?

Probably, inside a year. The kit treats it as a 6-12 month bridge and tells you so. The orchestration pattern (one job, one window, one result file) is reusable. The specific clipboard-paste trick might not be.

I am on Pro, not Max. Is this still worth $9.99?

If you have any cron firing claude -p, yes. The Pro credit is $20 and the audit usually shows that a single misconfigured cron drains it in a week. The trim checklist alone pays the kit back.

How does the 12-month update policy work?

Your purchase confirmation has a download link. Reuse that link within 12 months for the latest version. When Anthropic ships the next pricing change, the kit gets a new mitigation chapter and your existing download upgrades.

Secure checkout by Stripe. Instant download + guided Claude Code setup.