iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Research, Plan, Implement (RPI) is a useful way to organize Claude Code work: understand the codebase and task first, review a proposed approach before editing, then implement in small steps and verify the result. RPI is an editorial workflow, not a methodology Anthropic identifies as its own. Claude Code’s documentation does describe the underlying practices, including codebase exploration, planning, implementation, tests, and delegation to subagents.
What RPI means in a Claude Code workflow
RPI separates a coding task into three decisions: what is happening now, what should change, and how to make and check that change. This order helps keep an implementation grounded in the repository rather than in assumptions made from a brief prompt.
![Claude AI for Beginners Bible: [5 in 1] The Ultimate Guide to Automate Your Work, Save Hours Every Week, and Use AI for Real-World Results](https://m.media-amazon.com/images/I/415+fSJacsL._SL160_.jpg)
It is most useful when a task crosses files, has unclear behavior, or needs review before edits. A small, well-understood change may not need a formal plan or a subagent; extra coordination can cost more than it saves.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →Research the codebase and the actual problem
Begin broadly, then narrow the investigation as you learn where the relevant behavior lives. Anthropic’s common workflow guidance gives examples such as asking for a codebase overview, explaining its architecture patterns, or finding files that handle authentication.
#1 Best Overall
For a specific change, ask Claude Code to trace the relevant behavior and report the files and evidence it finds. Establish the project’s conventions, dependencies, expected behavior, and likely failure cases before settling on an implementation. Useful requests include:
- “Give me an overview of this codebase.”
- “Explain the main architecture patterns used here.”
- “Find the files that handle user authentication.”
- For a change request: ask it to trace the behavior involved, identify relevant files and conventions, and flag uncertainties or edge cases.
These are examples of prompts, not guaranteed search phrases or a promise that Claude Code will find every relevant file. Treat the findings as an investigation to check against the repository.
Plan before authorizing edits
For changes that warrant review, ask for a plan before implementation. A useful plan names the intended behavior, constraints, likely files, risks, and the checks that will establish whether the change worked. Review it as a proposal; correct misunderstandings before letting edits proceed.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Claude Code documents a plan permission mode. The CLI reference documents starting in that mode with --permission-mode plan; because command-line options can change, check the current CLI reference for the supported syntax and behavior. Anthropic’s workflow guidance also describes planning changes so they can be reviewed before touching disk.
Implement in bounded steps, then verify
Once the approach is agreed, ask Claude Code to make a scoped change rather than broad, unspecified edits. Small increments are easier to inspect and connect to a specific behavior. After implementation, ask it to run the relevant project tests, linters, or checks, and inspect the diff yourself.
- Make the agreed change. Keep the request tied to the reviewed plan and intended behavior.
- Run relevant checks. Choose tests or other project checks that exercise the changed behavior. Anthropic’s workflow examples include finding untested code, adding tests and edge cases, and running those tests.
- Inspect the outcome. Review the diff and the check results; ask for a summary of what changed, commands actually run, their observed outcomes, and unresolved risks.
A plan is not proof that the implementation is correct, and a claim that a check passed is not evidence unless the check was actually run and its result observed. Anthropic’s documentation presents verification as part of the workflow, not as a guarantee of correctness.
Rank #4
When to delegate research to a subagent
A Claude Code subagent is a specialized assistant with its own context window, instructions, tool access, and permissions. It can investigate a side task—such as mapping a subsystem—and return a summary, keeping lengthy exploration out of the main conversation. Its requests still count toward the same usage limits as requests in the main session, according to Anthropic’s subagent documentation.
Delegation is most useful when the investigation is separable, parallel, or likely to produce more logs, search results, or code excerpts than the main thread needs. For example, one subagent can map a subsystem while the main session examines another, or a specialist can perform a narrow review. Keep a small sequential task in the main session when it depends on continuous shared state or frequent decisions. Anthropic’s prompting guidance recommends subagents for parallel or isolated work and cautions against excessive delegation on straightforward tasks (prompting guidance).
Decide whether the task is worth delegating
| Consideration | A subagent is a better fit when… | Keep it in the main session when… |
|---|---|---|
| Independence | The subtask can proceed without frequent shared decisions. | Progress depends on ongoing choices with the main session. |
| Context load | Exploration may produce bulky logs, search results, or excerpts. | The work is compact and its details are needed throughout. |
| Parallel value | Distinct investigations can happen at the same time. | The steps are inherently sequential or concern one small file. |
| Tools and permissions | The work can safely use a narrower set of tools or permissions. | The task needs shared capabilities or context that is difficult to isolate. |
| Coordination and usage | A concise handoff is worth the coordination; subagent requests still count toward shared usage limits. | Summarizing, reconciling, and managing delegation would cost more than doing the task directly. |
Define subagent scope and configuration
Ask a subagent for a bounded deliverable, not an open-ended tour of the repository. Specify which area to investigate and request relevant file paths, observed behavior, uncertainties, and concise implementation implications. This makes its report easier to use in the main session.
Anthropic documents several places to define agents: managed settings for organization-wide definitions, .claude/agents/ for project-level definitions that can be version-controlled, ~/.claude/agents/ for user-level definitions, plugin directories for agents distributed with plugins, and CLI-defined agents for the current session. The documentation describes precedence among locations; use distinct names and check the documented precedence if definitions overlap.
The CLI reference documents --agents for session-defined subagents. Agent definitions include fields for the name, description, prompt, tools, and model. Make the description specific about when to use the agent, restrict its tools to those needed, and state the expected report in its instructions. These details and supported options are version-sensitive; consult the live subagent documentation and CLI reference when configuring them.
Recommended Free Tools
Put the sequence together
For a feature or bug fix, RPI can look like this: first ask Claude Code to map the relevant behavior and report the files, conventions, and unknowns; delegate an isolated investigation if it would materially reduce context load or allow useful parallel work; review a concrete plan that identifies the intended change and verification; then authorize a bounded implementation, run appropriate checks, and inspect the resulting diff and evidence.
The goal is not to add ceremony to every prompt. Use enough research and planning to make the change understandable and reviewable, delegate only when separation helps, and verify the repository’s actual outcome.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
