Execution-Grade Planning

Use this skill when a user wants to plan work deeply enough that implementer agents do not have to infer product context or make hidden product decisions.

Core principle

A strategic plan is not automatically implementer-proof. For product-critical work, convert broad plans into execution-grade packets before dispatching implementation agents. The planning agent and user should decide concepts, contracts, pass/fail semantics, safety rules, and verification oracles up front; implementer agents should primarily edit files, run gates, and report outcomes.

Trigger signals

Use this when:

• The user is worried implementers will infer the wrong context.
• The user says they are newer to coding and need conceptual/product correctness nailed down before code review.
• The task defines evals, product metrics, quality gates, safety boundaries, or product truth meters.
• A child plan or architecture doc is strong strategically but leaves schema, matching, failure precedence, setup safety, or report shape undecided.
• Multiple agents will execute sequentially from prompts.
• A one-PR plan hides several independent decisions.
• The user mentions existing repos or codebases that might overlap with what needs building. This is a signal to run Phase 0 audit before any planning — don't plan from scratch when pre-built components may already exist.