todo2code

Participant: codex (AI agent)

Understanding

AST parsing and Markdown chunking are deterministic but repeated for every run. Their cache keys must bind every input that can change output, while cached data must be treated as disposable acceleration rather than evidence.

Execution plan

  1. Map AST adapters, document chunking and output-directory boundaries.
  2. Add a shared versioned cache with atomic writes and fail-open recovery.
  3. Cache TypeScript per file, external adapters per source manifest and chunks per document.
  4. Prove cold/warm equivalence, invalidation, corruption recovery and provider isolation.
  5. Benchmark tracked snapshots, update repository evidence and publish main.

Blockers

Actual changes