feature-planning-artifacts

Feature Planning Artifacts

Create or update staged HumanLayer-style planning artifacts for work that needs research, design judgment, vertical slicing, or an executor-safe plan before implementation.

This skill is not a one-shot plan generator. It should normally take an input, delegate or perform reconnaissance, write or revise the current-stage artifact, then stop at the review gate. Do not continue to the next artifact until the current one is accepted or the user explicitly asks to proceed.

Treat detailed planning as meta-work: each artifact should make future implementation safer by defining success, surfacing decisions, turning recurring choices into policy when appropriate, and specifying evals/regression checks.

Open and resolved questions live in the design discussion. Plan iteration updates the plan itself.

Source Frame

Before writing or updating planning artifacts, load what applies:

• The selected strategic-roadmap item, opportunity card, architecture candidate, or NNN-*.md plan.
• Existing artifacts for this effort, especially the latest design discussion, outline, or plan.
• The coding-standards skill as the canonical source for design claims. Load the matching references; do not copy the standards into this skill or artifact.
• Project docs: AGENTS.md, README.md, CONTEXT.md, architecture docs, ADRs, verification config, current VCS state, and relevant existing plans.

The artifact rules in this skill are canonical. For source patterns and examples, read the vendored HumanLayer v2 planning skills at reference/humanlayer-riptide-v2-skills/ and the improve-codebase-architecture skill when architecture/deepening vocabulary matters.

Stage Contract

Default to the smallest stage that creates durable progress.

Stage	Use when	Inputs	Output	Stop gate
Research questions	The current-state unknowns are broad or ambiguous	Roadmap item, plan, user request	Questions about how the system works today	Stop after questions; ask whether to research
Research	Questions exist and current behavior must be established before design	Research questions or explicit scope	Descriptive current-state research	Stop after research; recommend design discussion
Design discussion	A feature or architecture direction needs judgment	Source item plus research or enough repo context	Options, recommendation, decisions, open questions	Stop for review; do not outline yet
Structure outline	The design is accepted or has one clearly recommended direction	Accepted/current design discussion	Vertical implementation slices	Stop for review; do not write final plan yet
Final plan	The outline is accepted or equivalent structure is provided	Accepted/current outline	Executor-safe implementation plan	Stop for execution handoff