planning
Break work into small atomic tasks with scope, risks, verification per task. Use after spec, before implementation.
SKILL.mdskills/core/planning/SKILL.md
GitHubplanning
A plan is a list of verifiable steps with owners and risks — not a paragraph of intent.
Inspection
Read the spec/requirements, dependency graph, risky unknowns, and prior similar tasks (what slipped last time).
Decision rules
Atomic tasks with independent verification; order by dependency; flag research/migration/security-review needs up front; size to single reviewable units.
Activate when
- After requirements/spec, before implementation; when decomposing epics or migrations.
Do NOT activate for
- Trivial single-file changes (plan inline); requirements gathering itself (see requirements-analysis).
Procedure
- Decompose into atomic tasks: each with scope, files touched, acceptance criteria, and how it will be verified.
- Order by dependency; surface risks, unknowns with owners, and rollback points for risky steps.
- Size tasks to single reviewable units; flag any task needing research, migration, or security review.
- Verify: every task independently completable and testable; plan re-read against the spec for coverage gaps.
Failure modes
Paragraph-of-intent plans; tasks without acceptance criteria; hidden dependencies discovered mid-implementation; no rollback points on risky steps.
Escalation
Requirements gaps → core/requirements-analysis; estimate disputes → ask with options.