writing-plans

Writing Plans

Part of Agent Skills™ by googleadsagent.ai™

Description

Writing Plans decomposes any implementation task into a sequence of atomic steps, each completable in 2-5 minutes, with exact file paths, complete code blocks, and verification commands. The plan is a self-contained execution script that another agent—or a future version of the same agent—can follow without needing the original conversation context.

The critical distinction between a plan and a TODO list is specificity. A plan does not say "update the API handler." It says "in src/api/routes/users.ts, add a GET /users/:id/preferences handler at line 47 that returns a UserPreferences object with fields theme, locale, and notifications, validated by the preferencesSchema defined in src/api/schemas.ts." Every step is unambiguous enough to execute mechanically.

Plans include verification steps after each implementation step: a test to run, a curl command to execute, or a UI state to confirm. This tight feedback loop ensures errors are caught immediately rather than compounding across subsequent steps.

Use When

• A task involves modifying 3+ files
• The implementation path is known but the execution is non-trivial
• Work needs to be handed off to another agent or resumed later
• The user asks for a plan, roadmap, or step-by-step breakdown
• You need to estimate effort before committing to implementation
• Complex refactoring requires precise ordering of changes

How It Works

graph TD
    A[Task Description] --> B[Identify Affected Files]
    B --> C[Determine Change Order]
    C --> D[Write Step 1: File + Code + Verify]
    D --> E[Write Step 2: File + Code + Verify]
    E --> F[Write Step N: File + Code + Verify]
    F --> G[Add Rollback Instructions]
    G --> H[Estimate Total Duration]
    H --> I[Present Plan for Approval]
    I -->|Approved| J[Save as plan.md]
    I -->|Needs Revision| B