Plan Smarter: Code-Informed Plans for AI Agents

Where this fits

You ask your coding agent to refactor a service into smaller modules. Instead of a generic checklist, it returns a step-by-step plan that references the actual files, functions, and test suites in your repo. That's the gap sesori-plan-maker fills: turning a high-level goal into an actionable, code-grounded implementation plan without building a heavyweight planning system.

Why agents benefit

• Grounded in your codebase – The skill reads the current project structure and uses real symbols, paths, and dependencies to shape each step, so plans aren't abstract.
• Proportional effort – It prefers a short, useful plan over an elaborate process. For a small change, it won't over-engineer; for a large refactor, it expands appropriately.
• Explicit self-invocation rules – The agent knows when to plan (user asks, or during large multi-step efforts) and when to stay silent (routine fixes), reducing noise and wasted tokens.
• Respects user authority – If you ask it to do something outside planning, it briefly confirms and then proceeds without nagging, keeping the workflow smooth.
• Plan updates become first-class – When requirements shift, the agent can update the existing plan rather than starting from scratch, maintaining a living tracker.

Practical scenarios

Breaking up a monolith into services

You tell the agent: "Split the payment module into a separate service." The skill reads the current module, identifies internal dependencies, and produces a phased plan: extract interfaces, move code, update imports, add tests, then deploy. Each step references specific files and expected PR splits.

Introducing a new feature with multiple PRs

You're adding user authentication to an existing app. The agent plans the work across three PRs: database schema changes, backend API endpoints, and frontend integration. The tracker lists acceptance criteria per PR and checks off completed items as you go.

Refactoring a legacy function without breaking tests

You ask for a plan to modernize a 500-line function. The agent examines the function's call sites, existing test coverage, and edge cases, then proposes incremental changes with verification steps at each stage. The plan includes rollback points and risk notes.

Add it to your agent workflow

Install the skill with:

npx skills add sesori-ai/sesori_apps_monorepo --skill sesori-plan-maker

Then, in your agent conversation, simply say: "Make a plan to refactor the auth module into separate handlers." The agent will load the skill, inspect your code, and produce a structured plan. You can later ask: "Update the plan to include rate limiting" and it will adjust the tracker accordingly.

Compare before adopting

Before integrating sesori-plan-maker, evaluate other planning approaches: generic prompt templates, other agent skills like plan-and-execute, or manual tracking in issue boards. Compare quality signals like how well the plan reflects your actual codebase, maintenance freshness (last commit, open issues), and whether the skill's philosophy matches your workflow. Also consider if you need a full tracker or just a one-off plan.

Why it is worth tracking

With 105 stars and an active repo, sesori-plan-maker shows early community interest. Its focus on proportional planning and code-awareness addresses a common pain point: agents that over-plan or under-plan. Watch for updates that improve dependency analysis or add integration with common issue trackers. Re-evaluate when your projects grow in complexity or you start using multiple agents that need consistent planning behavior. For more details, see the skill page on OpenAgentSkill.