Skip to content

02 - The Core Loop

A SpecLoop run has a stable shape:

  1. Load cold context.
  2. Load hot run context.
  3. Check guardrails.
  4. Execute or receive runner evidence.
  5. Compare evidence with specs.
  6. Write findings immediately.
  7. Render reports.
  8. Suggest regression tests.

Agentic browser runs can fail halfway through. Writing each finding immediately keeps evidence durable and makes the run resumable.

SpecLoop should distinguish:

  • pass: behavior matches spec;
  • fail: behavior violates spec;
  • drift: product behavior changed and the spec may need review.