Pincer Planning
Productivitypincer-planningPincer Planning is an Agent Skill for productivity published in the majiayu000/claude-skill-registry repository on GitHub: a SKILL.md instruction file an AI agent can load to do this task.
Bidirectional planning for ambiguous or design-heavy tasks. Use when the user requests help breaking down a problem or formulating a plan, and also when an existing plan has too many gaps or needs refinement; converge top-down framing with bottom-up inventory and explicit questions.
How to use this skill
Bring this guide into your coding agent with a prompt tailored to the tool you use.
- Open your project in Codex.
- Copy and paste the prompt below.
- Review files and risks before approving any action.
Inspect the Agent Skill at https://github.com/majiayu000/claude-skill-registry/blob/HEAD/skills/data/pincer-planning/SKILL.md and all supporting files. Report provenance, permissions, scripts, dependencies, and security risks. Propose an exact installation into .agents/skills/pincer-planning, but wait for my approval before writing files or running scripts.
Copying this prompt does not install or run the skill. Review third-party files before use.
Pincer Planning
Overview
Create a plan by pairing a top-down framing with a bottom-up inventory, stress-testing both with questions, then merging into a single, gap-checked plan. Use for design, strategy, or fuzzy scopes.
Quick start
- Draft top-down framing (goal, constraints, success criteria).
- Draft bottom-up inventory (facts, assets, risks, unknowns).
- Ask clarifying questions (use AskUser).
- Flip perspective and refine the opposite view.
- Merge into a single plan and list remaining gaps.
Core Guidance
- Start with either top-down or bottom-up; explicitly label the other side as provisional.
- Ask 3–7 targeted questions that would change the plan, then wait for answers.
- After answering, re-run the opposite perspective to find contradictions or gaps.
- Converge: produce a merged plan plus a short gap list (what’s still unknown).
- Keep planning separate from execution; do not run tools during planning.
Trust / Permissions
- Always: Read local files, ask clarifying questions, draft plans.
- Ask: Any writes to disk (e.g., saving plan.json), network calls, or execution.
- Never: Destructive actions or credential exfiltration.
Resources
references/pincer-checklist.md: Question prompts and convergence checks.templates/plan.json: Plan scaffold with top-down and bottom-up sections.
Validation
- Confirm top-down and bottom-up sections converge; list explicit gaps.
- Ensure questions are answered before final plan is delivered.