Skip to content

code_review.md

Latest commit

 

History

History
46 lines (34 loc) · 1.43 KB

code_review.md

File metadata and controls

46 lines (34 loc) · 1.43 KB
Raw

Code review

Read AGENTS.md and the nearest subtree AGENTS.md first. Lead with concrete findings and file refs.

Look for

  • Correctness against the requested behavior
  • Regression risk: CLI, schemas, workflows, runtime state
  • Checks: bun run agent:verify ran, with exact results and skipped optional checks
  • Scope: no unrelated refactors, generated files, caches, or local runtime artifacts
  • Secrets, transcripts, credentials, screenshots, sensitive local paths
  • For Lens/Waygent: saved files, event schemas, non-blocking behavior

Lens

  • JSON is the saved record; SQLite is a cache you can rebuild
  • Internal Lens failures must not change the wrapped command's exit code
  • Schema changes are additive unless a versioned migration is designed
  • Dashboard/API type drift: bun run typecheck plus contract, view, and consumer tests when relevant

Runtime

  • Providers do not write SQLite or Lens directly
  • Scheduler changes respect safe task groups, dependency checkpoints, and failure barriers
  • Recovery stops on missing handlers or human-decision classes. No fake progress.
  • Behavior changes need targeted tests or scenario coverage

Docs

  • Local Markdown targets resolve without the network
  • Required live proof stays optional and is reported separately

Output

  1. Findings first, by severity
  2. File and line refs
  3. Open questions
  4. Short check summary

If there are no findings, say so, and name leftover risk.