Reasoning Audit

Find why your
agent fails

Your AI agent fails silently. Find out why in 48 hours for $49. Send your agent prompts, code, and one example failure. You get back a Google Meet walkthrough plus a 2-page Markdown report identifying your top three failure modes with concrete fixes.

Refund if no fix in 14 days
Reasoning Audit

How it runs

48 Hours. Concrete Fixes.

01

Intake

Email to book. Send your agent prompts, code, and one example failure.

02

Review

I review your stack and identify the failure modes within 48 hours.

03

Walkthrough

A 20-30 min Google Meet plus a 2-page report detailing ranked fixes.

04

Guarantee

Implement the fixes. If no fix lands within 14 days, you get a full refund.

The Artifacts

See exactly what you pay for.

sample-audit.md

An anonymised real audit deliverable. Identifies the top 3 failure modes with concrete code fixes.


### #1 — Silent tool-call failure on Zendesk post

**What's happening:**
The agent calls zendesk.tickets.create() inside a try/await. When Zendesk returns 5xx, the call throws, the catch logs to stderr, the agent enters a success branch.

**Fix:**
Replace bare try/await with structured retry and a postcondition asserting the ticket actually exists.

silent-tool-call-contract.md

A proof-of-concept LOGIC.md contract that makes the most common failure mode structurally impossible.


steps:
  - id: create_ticket
    retry: { max: 3, on: [HTTP_5XX] }

quality_gates:
  - id: no_silent_failure
    assert: verify_ticket_exists.passed == true
    on_fail: { action: distinct_failure_branch }