Plan Implementation

Plan Implementation

Use this skill when a change is too large, risky, or ambiguous to implement directly.

Native Integration

Use the active agent's native planning mechanism first, such as Codex subagents, Claude Plan mode, OpenCode Plan agent, or Cursor Ask/Plan/custom modes. This skill defines the content and acceptance criteria for the plan; it is not a replacement for native read-only planning controls.

Workflow

1. Inspect the system.

   • Read the relevant source, tests, docs, config, and existing patterns.
   • Identify public interfaces, ownership boundaries, and feedback loops.
   • Check CONTEXT.md and any ADRs when they exist.

2. Define the target behavior.

   • State the current behavior and desired behavior.
   • List in-scope and out-of-scope changes.
   • Name affected users, callers, commands, APIs, files, or workflows only when needed for implementation clarity.

3. Choose the approach.

   • Prefer existing patterns over new abstractions.
   • Keep changes as small as the goal allows.
   • Define any public API, schema, CLI, UI, or file format changes precisely.
   • Record assumptions where the repo or user did not provide a fact.

4. Plan verification.

   • Identify the highest useful test seam.
   • Include manual checks only when automation cannot cover the behavior.
   • Include regression coverage for bugs.

5. Produce the plan.

   • Keep it concise but decision-complete.
   • Another engineer or agent should be able to implement without making product or architecture decisions.

Plan Shape

# Title

## Summary
- What will change and why.

## Implementation
- Behavior-level edits grouped by subsystem.
- Public interfaces or contracts that change.

## Test Plan
- Automated checks.
- Manual checks, if needed.

## Assumptions
- Defaults chosen because the repo or user did not specify them.

Done When

• The plan can be implemented without unresolved decisions.
• The plan names verification commands or scenarios.
• Assumptions and risks are visible.