What problem does it solve? Agent-generated plans often fail at two altitudes: they are either too vague ("add caching", "handle auth") or lost in low-level detail with no product vision, and they rely on training-cutoff knowledge that may be outdated. This Skill performs an adversarial senior-engineer review grounded in live codebase research and current web sources, then rewrites the plan into a scoped, evidence-backed version. ## Core Features & Use Cases - Altitude Diagnosis: Classifies each plan section as fog (too vague), tunnel (too granular), or mixed using concrete tests like the start-tomorrow test and a vague-word blacklist. - Evidence-Based Adversarial Review: Every finding cites a file and line or a dated web source, with blocker/major/minor severities and concrete fixes; unverifiable claims are labeled [training-data, unverified]. - Plan Promotion: Produces a rewritten plan with goals and non-goals, versioned decisions with rejected alternatives, sequencing with verifiable milestones, rollback strategy, and open questions for humans. - Use Case: Before committing to an agent-written migration plan, run a senior review that checks pinned dependency versions in the lockfile, searches current release notes for deprecations, flags a missing data migration as a blocker, and returns a promoted v2 plan. ## Quick Start Ask the assistant to review the current plan like a staff engineer using the junior-to-senior skill before implementing it.
