README-driven development pairs cleanly with Claude Code's TDD loop

Source type	Practitioner
Confidence	Medium
Last verified	2026-04-07
Topics	claude-code

Insight: Willison's three-release run on scan-for-secrets shows a workable pattern: write a precise README first, then drive Claude Code through a red/green TDD loop against that spec. The README functions as the executable specification, and the model fills in tests and implementation around it. Successive versions added encoding-aware detection (literal, backslash-escaped, JSON-escaped), streaming output, multi-file inputs, and a Python API across a few hours of iteration.

Detail: This corroborates the broader 'spec-first' pattern already in the knowledge base: when intent is captured concretely up front, agentic iteration becomes cheap and the human stays in the architect role rather than reviewing every diff.

Sources

• Simon Willison — "scan-for-secrets v0.1" (2026-04-05)
• Simon Willison — "scan-for-secrets v0.1.1" (2026-04-05)
• Simon Willison — "scan-for-secrets v0.2" (2026-04-05)
• Simon Willison — "scan-for-secrets follow-up" (2026-04-06)