| name | deep-plan |
| description | Produces a deep implementation plan with clarifications, multiple solution variants, and adversarial critique. Use when the user asks for architecture, a serious implementation plan, design options, or a high-confidence roadmap before coding. |
Deep Plan
Use this skill for non-trivial planning.
Canonical Planning Inputs
Before planning, load the relevant role prompts:
references/codebase-analyzer.md
references/architecture-planner.md
Use them as the baseline for codebase mapping and implementation-plan quality.
When to Ask Questions First
Ask clarifying questions before planning if any of these are unclear:
- success criteria
- scope boundaries
- target users or systems
- performance or security constraints
- backward-compatibility expectations
- whether the task is conceptual or tied to a specific repository
If the task is already well-scoped, proceed without blocking the user unnecessarily.
Workflow
- Restate the request in a clearer, stricter form.
- Identify ambiguities, assumptions, and hidden constraints.
- If this is repository-specific, analyze the codebase first using the standards from
references/codebase-analyzer.md.
- Build a concrete implementation plan using the standards from
references/architecture-planner.md.
- Generate at least 3 viable solution variants.
- Keep the variants genuinely different in tradeoffs, not just reworded copies.
- Critique each variant adversarially:
- logical gaps
- hidden assumptions
- maintainability risks
- security and reliability concerns
- ethical or product risks when relevant
- If a stronger external reasoning model or review tool is available, use it for an additional critique pass; otherwise do the critique yourself explicitly.
- Produce a final recommendation with rationale and a comparative rating.
Output Structure
Return a structured report with:
- improved task statement
- clarifying questions or explicit assumptions
- constraints and acceptance criteria
- codebase context if applicable
- step-by-step implementation plan
- variant A / B / C
- critique per variant
- final recommendation
- residual risks and open questions
Constraints
- Do not implement code.
- Prefer executable plans over abstract essays.
- State uncertainty explicitly.
- Respond in the user's language.