deepen-feature-plan_skill

Overview

This skill critiques and stress-tests an existing feature plan located in ~/.ai/plans. I act as a senior engineer: hunting for gaps, risky assumptions, and mismatches between the plan and the actual codebase. Use this to surface problems early so implementation avoids costly rework.

How this skill works

I read the plan and compare its claims to the codebase: file locations, patterns, existing services, and similar features. I flag specification gaps, missing requirements, architectural misfits, hidden complexity, blast radius, and edge cases. When details are vague I push with direct, actionable questions to force concrete acceptance criteria and testable verification steps.

When to use it

• After a plan already exists in ~/.ai/plans and before implementation starts
• When a plan contains vague directives ("follow existing pattern", "handle errors appropriately")
• If the plan adds new abstractions, schema changes, or external integrations
• When you need a risk assessment and a clear rollback/testing strategy
• Before scheduling work that touches core systems or many files

Best practices

• Treat every plan as incomplete; expect blind spots and ask for specifics
• Force measurable acceptance criteria: inputs, outputs, error behavior, performance targets
• Verify proposed file paths, function names, and patterns against the repo rather than assuming locations
• Prefer minimal changes that match existing patterns; only add new abstractions with clear justification
• Define testing: unit, integration, mocks for external services, and end-to-end verification steps

Example use cases

• Deepen a plan that says "modify search.ts" by identifying the exact function and dependent callers
• Critique a schema change plan by listing migration, backward-compatibility, and rollback steps
• Assess a dual-write proposal by comparing with existing multi-provider patterns and suggesting simpler feature-flag options
• Find missing observability: add metrics, logs, and alerts the plan didn't mention
• Turn vague acceptance criteria into concrete test cases and a verification checklist