ada-pre-implementation-audit

Extract the REQs relevant to the planned scope with their claimed statuses, then confirm each claim with grep or read_file against source and test files. Build a discrepancy table showing what traceability says versus what the code actually contains before writing the plan.

Use delegate_task batch mode with three agents covering independent module clusters such as domain, services, and components. Each agent reads full file paths, notes REQ references and TODOs, and reports test coverage; cross-check all findings in the main context before trusting them.

Skip it for greenfield projects with no existing code or traceability dependencies, plans that do not reference documented statuses, or when a full project audit was just completed in the same session. Re-auditing immediately is redundant.

Yes, the post-plan feasibility review mode treats every claimed fact in the plan as a hypothesis and verifies it against source with file:line evidence. It checks caller blast radius, test instantiation patterns, and CI ordering, then returns a feasible, revise, or infeasible verdict.

Traceability matrices are lagging indicators: statuses go stale within days and file lists miss newly added modules. Plans built on them schedule work that is already done or miss real gaps, so claims must be verified against code first.