What problem does it solve? After a dev-story or code-review phase finishes, task completion checkboxes are often marked blindly without reading the actual code, and review stages rubber-stamp dev results without independent audits. This Skill enforces evidence-based verification of every task with file:line proof and writes the validated results back to the story database. ## Core Features & Use Cases - Evidence-Based Task Verification: Reads actual source files and requires file:line evidence before any task can be marked complete, with per-task-type validation standards for code, tests, migrations, docs, and DB fields. - Two-Stage Dev/Review Differentiation: Forces the review phase to independently re-audit every dev marker, upgrading verified items to double-check marks with R-verified timestamps or downgrading mismatches to failures with CR findings, and blocks copy-paste via SHA-256 comparison. - Structured DB Backfill: Assembles verified tasks as Markdown strings with strict marker placement rules and merges them into the stories table via upsert scripts, including file_list tables and semantically defined test_count values. - Use Case: A code-review agent finishing a story review invokes this Skill to re-read each claimed implementation, confirm the evidence still matches, upgrade markers with timestamps, and write the audited task list back to the database. ## Quick Start Ask the agent to run the tasks-backfill-verify skill to independently verify each story task against the code and backfill the results into the database.
