agent-plan
by 2270525352
Plan a new project end-to-end into an AI-readable tree so a single AI window (one Claude or one Codex window) executes by following marked tasks instead of re-deciding what to build. Use when the user wants requirements aligned and frozen before coding, the user's exact words preserved verbatim as the alignment anchor, granular tasks with execution feedback, Claude/Codex /goal automation prompts, scheduled drift testing (Claude Code cron or Codex App) that runs tests and re-checks the source of truth, review gates, strict git checkpoint discipline, or deterministic guardrails (Claude Code hooks + git hooks) that enforce task scope and commit discipline.