What problem does it solve? Code that passes tests, lint, and CI can still ship stubs, silently break existing callers, contain concurrency or security defects, or rest on fake test evidence. This Skill provides a disciplined review method that catches exactly the failures mechanical gates miss, before a change is committed or merged. ## Core Features & Use Cases - Four-question review framework: verifies the change delivers what it claims (stub and wiring detection), does not break existing consumers (contract-surface analysis), is correct under real inputs (defect catalogue covering concurrency, resources, boundaries, async, errors, security), and is backed by honest test evidence with no hidden debt. - Plan-backed and planless modes: reviews against explicit Plan truths and design constraints when available, or against the change's own stated intent for bare diffs, commits, and commit ranges. - Structured verdicts and reports: emits PASS / REVISE / BLOCK verdicts with Blocker, Warning, and Info findings, each citing file, line, evidence, impact, and fix direction, plus a detailed rubric with probe commands and worked examples. - Use Case: Before merging a pull request, ask for a review of the working tree changes; the Skill reads the diff once, probes consumers of every changed symbol with ripgrep, and returns a report flagging a stubbed API handler as a Blocker and an undocumented contract change as a Warning. ## Quick Start Ask the assistant to review the current working tree changes or a specific commit using the df-implement-review skill before you commit or merge.
