plan-with-research
Build step-by-step plans grounded in evidence and research. Use when the user asks for a plan, roadmap, steps, or strategy for code, data, or pipeline work.
When & Why to Use This Skill
The 'plan-with-research' Claude skill is designed to generate high-precision, evidence-based technical roadmaps and strategic plans. It bridges the gap between high-level goals and execution by inspecting local codebases and data artifacts to ensure every step is grounded in reality. This skill is essential for developers and data engineers who require structured workflows, risk assessment, and validation-driven planning for complex software or data pipeline tasks.
Use Cases
- Technical Roadmap Development: Create a 3-7 step implementation plan for new features or system refactoring, using repository evidence to justify each phase.
- Data Pipeline Strategy: Design robust data processing workflows that include explicit validation checks and success criteria based on existing data artifacts.
- Risk Mitigation & Rollback Planning: Formulate complex deployment strategies that identify potential dependencies and risks, providing clear fallback options to ensure system stability.
- Evidence-Based Decision Making: Use the 'Honesty Guardrails' to distinguish between proven facts and assumptions when planning architectural changes or infrastructure migrations.
| name | plan-with-research |
|---|---|
| description | Build step-by-step plans grounded in evidence and research. Use when the user asks for a plan, roadmap, steps, or strategy for code, data, or pipeline work. |
Plan With Research
Overview
Create concise, evidence-backed plans with explicit validation for each step.
Workflow
- Clarify goal, success criteria, constraints, and timeline.
- Inspect the repo or data artifacts to confirm reality (prefer local files and logs).
- Research gaps in docs or external sources only if needed.
- Draft a plan with 3-7 steps, dependencies, and risks.
- Include validation checks and rollback or fallback options.
Plan Format
- Step: verb + concrete deliverable.
- Evidence: cite repo files, docs, or experiments.
- Validation: tests, dry runs, or metrics.
Honesty Guardrails
- Separate evidence from assumptions; label unknowns explicitly.
- If confidence is low, say so and propose the smallest check to raise it.
Gating
- If evidence is missing, state assumptions and ask one targeted question.
Acceptance Criteria
- Each step has a deliverable, evidence note, and validation check.
- Research sources are listed or explicitly noted as not used.
- Risks and rollback/fallback are included.