Context bloat
Tracked lockfiles, generated folders, large blobs, and missing ignore rules that pull agents away from useful files.
What the preview catches
Discovery churn
Repo shapes that make agents repeatedly grep, read, and rediscover the same files before changing code.
Workflow gaps
Missing agent instructions, eval harnesses, or repo maps that leave every new session starting cold.
CI threshold
Use sravan27/context-os@v2.9.0 or the local checker with --max-score to stop new generated noise or missing agent docs before they land.
Drop it into CI
The checker is now a pinned GitHub Action with a green smoke test in context-os CI. Start at a forgiving threshold, then lower it after adding repo guidance, ignore rules, and validation notes.
name: Agent cost leak check
on: [pull_request]
jobs:
agent-cost-leak:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- uses: sravan27/context-os@v2.9.0
with:
max-score: "40"Private repo audit — pick a tier
The free scan above reads only your public file tree. The paid audit looks at what actually drives agent waste: real repo structure, your CLAUDE.md/agent instructions, ignore rules, and workflow habits — then ships something you can act on. Manual work, one repo at a time.
Diagnostic Fix
$500
- Async repo + workflow review
- Root-cause diagnosis of the biggest cost/context leak
- One focused fix or implementation recommendation
- Short handoff note with next steps
Recommended
48-Hour Audit + Fix Plan
$1,000
- Everything in Diagnostic, plus:
- Full audit: structure, CLAUDE.md/agent docs, ignore rules, discovery churn
- Written report with prioritized findings
- A pinned GitHub Action cost-leak gate, tuned to your threshold
- One concrete fix path implemented or specified
- One revision pass inside the original scope
Audit + Handoff
$1,500
- Everything in the 48-Hour Audit, plus:
- Architecture / context map for your repo
- Operator runbook your team can reuse
- Two revision passes inside scope
- Optional white-label delivery for agencies
Payment is handled by service-appropriate invoice after scope agreement.
Why trust this
A public, reproducible benchmark
The method behind this audit ships as context-os (MIT). On a 36-call A/B against identical fixtures it measured −40.9% total tokens and −35.3% wall-clock (Sonnet, 6/6 prompt-level wins, paired t-test p=5.1e-7). Raw data · methodology. Your repo will differ — the audit measures yours specifically.
See the deliverable first
A clearly-labeled sample audit report shows exactly what you get. No mystery, no fake testimonials.
The same gate runs in CI
The cost-leak check is a pinned GitHub Action with a green smoke test in context-os CI. The audit tunes it to your repo so regressions get caught on every PR.
What happens after you pay
- You get a kickoff email asking for repo access (or a clone), one sample task/transcript, and your goal.
- The 48-hour delivery window starts once access and context arrive — not at the moment of payment.
- You receive the report, the tuned CI gate, and the fix path, plus one revision inside scope.
Limits & refund
One repo, one primary workflow per audit. This reduces and catches agent cost/context waste; it does not raise provider rate limits or guarantee a specific dollar saving for your repo. If, after intake, I don't believe I can deliver useful findings for your codebase, I'll say so and refund in full before any work starts.
Independent service. Not affiliated with, endorsed by, or sponsored by Anthropic, Cursor, or OpenAI. “Claude”, “Cursor”, and “Codex” are trademarks of their respective owners.