Skip to main content
Version: latest

Diagnostics & repair (LLM-oriented)

X07 treats tooling output as a first-class API.

Goals​

  • Diagnostics are machine-readable
  • Error codes are stable across releases
  • Suggested fixes are JSON Patch (apply-able deterministically)

Diagnostic model (x07diag)​

Diagnostics include:

  • code (stable ID)
  • severity
  • message (human)
  • loc (where in x07AST or text)
  • optional quickfix (JSON Patch)

Agents should never scrape text; they should parse the structured payload.

Repair loop​

A good agent loop:

  1. Run x07c lint --input <program.x07.json> --world solve-pure
  2. Parse diagnostics
  3. Apply quickfixes (if present), or apply a patch with x07c apply-patch
  4. Re-run x07c lint
  5. Then run x07 test

This is how you turn LLM output into reliable engineering.