Forma Plan

---
name: "forma-plan"
description: "Clarify goals, constraints, boundaries, and acceptance criteria."
---

# Forma Plan

Clarify goals, constraints, boundaries, and acceptance criteria.

## Interaction Semantics

- Use this skill to converge an executable task contract with the user before any repo plan files are written.
- Keep facts and decisions separate: grounding producers confirm discoverable facts, while the user owns choices that change product or engineering semantics.
- Stay in clarification or proposal mode until the user confirms the proposal is ready to hand off.
- Keep this skill chat-only: it produces clarification, a proposal, or a handoff; execution work and planning files belong to later stages.

## Mode Check

- First determine whether the current collaboration mode is plan-oriented.
- If the current agent is not in plan mode, stop and tell the user to switch to plan mode before continuing.
- Treat mode switching as host-controlled; continue only after the host environment confirms plan mode.

## Workflow

- Load bundled planning references from the current triggered skill package.
- Always load `references/output-format.md` and `references/plan-stage-rules.md` before deciding whether to clarify, block, propose, or hand off.
- Classify each unresolved item as a discoverable fact or a user-owned decision, then record the decision dependencies. Ask a decision separately when another decision depends on its answer or when its impact needs focused review; group independent decisions from the same active frontier when their answers cannot invalidate or reframe one another.
- Include the recommended answer and rationale for every decision question. Use no fixed question limit: choose the question group from the actual dependencies and coherence of the current task, wait for the user's feedback, then recompute the next decision frontier from that feedback.
- If the task intent is not yet specific enough to bound repository or domain research, continue clarification instead of exploring the repository or writing repo files.
- Once the Intent-Ready Gate passes, hand off any blocking fact questions to the selected grounding producer before continuing toward `proposal-ready`. Include the settled intent, exact fact questions, decisions waiting on those facts, and `Return To: plan`.
- After grounding returns, resume the plan stage, use confirmed facts to resolve the waiting decisions, and repeat the clarification-grounding loop only when another execution-shaping fact gap remains.
- Classify the issue as `Plan Strategy: step-execution`, `loop-exploration`, or `hybrid`; for loop-exploration, hybrid, generated-artifact, import/export, migration, batch-processing, formal/destructive write, or evidence-producing plans, settle concrete artifact paths, evidence paths, validation gates, and proof requirements before proposal-ready.
- Load `references/profile-standards.md` before `proposal-ready`. When it declares project standards, select only those that materially change this task's scope, approach, constraints, acceptance, validation, proof, or stop conditions, and show how each selected standard will be enforced in the task contract.
- For behavior, architecture, ownership, or integration work, load `references/engineering-quality-model.md`, identify the dimensions that materially plan success, and include their project-specific criteria and proof expectations in the proposal.
- Enter `proposal-ready` only after the Proposal-Ready Gate passes: no required grounding producer remains unrun, all execution-shaping decisions are resolved, and the executor will not need to choose product or release semantics.
- Keep the proposal in chat only. The finalize stage writes `plan.md` and `tasks.md`; workflow scripts are outside plan-stage operation.
- After the user reviews and confirms the proposal, stop and hand off to the finalize stage to write `plan.md` and `tasks.md`.

## Always Load

- `references/output-format.md`
- `references/plan-stage-rules.md`
- `references/proposal-decision-gate.md`
- `references/grounding-handoff.md`
- `references/profile-standards.md`

## Load As Needed

- `references/engineering-quality-model.md`

## Requirements

- Treat the plan stage as the convergence step before lock, not as a plan-file-writing skill.
- When loading bundled planning references, resolve them relative to the current triggered skill package.
- Treat user-provided source references as planning context only after their relevant contents are available in the current session or through an explicitly injected/profile-owned source adapter.
- Source-context tools such as GitHub, `gh`, or network access are available only when an injected/profile-owned source adapter provides them. Otherwise ask the user to paste the authoritative source material needed for planning.
- While the user context is still incomplete, stay in clarification mode. Ask only the active user-owned decision frontier needed to converge Goal, Scope, Approach, and the validation model for tasks, shared checks, and final issue closure; do not mix dependent decisions merely to reduce turns.
- Before the user confirms that the context is sufficient, stay in conversation-level clarification. Repository exploration belongs to grounding; plan files and `forma-workflow.sh` belong to lock.
- Load and follow `references/proposal-decision-gate.md` for the proposal decision gate.
- Load and follow `references/grounding-handoff.md` for grounding handoff selection.

## Output

- Follow `references/output-format.md`.