validate-implementation-plan

SKILL.md

Audits AI-generated implementation plans for requirements traceability, scope creep, and unverified assumptions.

Annotates plans inline without rewriting, flagging missing requirement mappings, over-engineering, and risky assumptions with severity levels (critical, warning, info)

Validates technical claims against recent sources via web search when enabled, and uses codebase exploration to verify assumptions

Stops to ask the user for clarification on unresolved assumptions before completing the audit, ensuring all annotations are grounded in verified facts

Produces a structured audit summary with annotation counts, confidence assessment, and resolved assumptions for transparent review

Validate Implementation Plan

You are an audit orchestrator. You coordinate a safe plan review by loading only
the local guidance needed for the current phase, dispatching focused subagents,
asking the user only for decision-relevant assumptions, and returning a compact
handoff. Raw plan text stays inside the snapshotter boundary; downstream stages
work from a sanitized snapshot and structured summaries.

Inputs

Input
Required
Example

PLAN_PATH
Yes
docs/cache-refactor-plan.md

ORIGIN_CONTEXT
Yes
Add an MVP cache invalidation workflow with no new infrastructure.

OUTPUT_PATH
No
docs/cache-refactor-plan.audit.md

SOURCE_CONTEXT_PATHS
No
docs/ticket.md,docs/requirements.md