octocode-plan_skill

Overview

This skill plans multi-step engineering work by orchestrating evidence-driven research and producing actionable implementation plans. It enforces a strict Understand → Research → Plan → Implement → Verify flow and never guesses—every recommendation must be validated. It integrates with local and external research tools and generates traceable plans with explicit file paths and references.

How this skill works

The agent first clarifies scope, constraints, and desired mode (interactive vs auto) before delegating research to specialized skills: a local code search for workspace questions and an external research skill for public repos and docs. It synthesizes findings with confidence levels, drafts a step-by-step implementation plan that references evidence, and requires explicit user approval before any code edits. Verification steps and risk mitigations are included for each plan.

When to use it

• When you ask to “plan this feature”, “plan refactor”, or “research & plan”
• When changes span multiple files, systems, or repositories
• When you need evidence-backed choices (patterns, libraries, APIs) before coding
• When there’s risk of breaking existing functionality or unclear design trade-offs
• When you want explicit task tracking and traceability from research to implementation

Best practices

• Always run Phase 0: confirm scope and constraints before any research
• Delegate local code questions to the local search skill and external questions to the research skill
• Require at least two sources or an authoritative primary source for key decisions
• Write clear plan steps with full file paths and direct references to research findings
• Hold a mandatory approval gate before implementing any plan steps

Example use cases

• Plan an auth refactor: discover local auth flows, research best practices, produce stepwise migration plan with tests
• Research & plan API versioning: gather local endpoints and external docs, produce backward-compatibility steps and rollout strategy
• Plan a feature spanning frontend and backend: spawn parallel research tasks, synthesize into a unified implementation plan
• Refactor a module safely: map usages, propose minimal edits, list pre- and post-conditions and validation steps