RPI—Research, Plan, Implement—is a useful way to organize Claude Code work: investigate the codebase first, agree on an approach before editing, then make and verify a bounded change. It is an editorial workflow, not a methodology Anthropic identifies as its own. Claude Code’s documentation does, however, describe the component practices: codebase exploration, planning, implementation, verification, and selective delegation to subagents.
What RPI means in a Claude Code workflow
RPI is a sequence for reducing avoidable guesswork. Research establishes what the repository does and where a change belongs. Planning turns that evidence into a reviewable proposal. Implementation applies the agreed change, followed by checks against the actual codebase.
![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)
As an Amazon Associate I earn from qualifying purchases.
The sequence is flexible rather than ceremonial. A small, obvious edit may not need a formal plan or delegation. A change spanning unfamiliar components, carrying behavioral risk, or requiring substantial investigation benefits more from distinct research, planning, and implementation stages. Anthropic’s common workflow guidance describes examples of exploring code, planning changes, applying edits, and verifying work.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Research the repository before choosing a fix
Begin broadly, then narrow the investigation to the behavior at issue. Anthropic’s workflow examples include asking Claude Code for an overview of a codebase, its architecture patterns, or the files that handle a particular feature. For a specific change, follow up by asking it to trace the relevant behavior and identify conventions, dependencies, and likely failure cases.
#1 Best Overall
Ask for evidence you can inspect: relevant file paths, how a request or data flow moves through the system, existing tests, and any uncertainty. That makes it easier to distinguish an observed repository fact from an assumption. A useful starting sequence might be:
-
“Give me an overview of this codebase.”
-
“Explain the main architecture patterns used here.”
-
“Find the files that handle user authentication, and trace how a login request is validated.”
Recommended Free Tools
Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy. -
“Which conventions and tests should a change to this behavior follow? Point to the relevant files and note anything you could not confirm.”
These are adapted from examples in Anthropic’s documentation, not tested prompt formulas. Their value is in moving from orientation to a focused, evidence-backed question.
When to delegate research
Use a subagent when exploration is separable from the main task, can proceed independently, or would otherwise fill the main conversation with search output, logs, or code excerpts. For example, one agent can map authentication while the main session examines the API boundary; a narrow review can also be delegated after an implementation.
Give the subagent a bounded task and specify the report you need: relevant paths, observed behavior, uncertainties, and implications for implementation. Subagents have their own context, instructions, tool access, and permissions, and can return a summary so the main conversation stays focused. Their requests still count toward the same usage limits as the main conversation. See Anthropic’s subagent documentation.
Plan before editing when review has value
Before approving changes, ask Claude Code for a concrete proposal: intended behavior, constraints, files likely to change, risks, and the checks that would show whether the work succeeded. Review whether the plan preserves existing behavior and whether its verification steps cover the affected paths. If the investigation uncovers a different architecture or a hidden dependency, revise the plan rather than letting an early guess dictate the implementation.
Claude Code’s CLI reference documents the --permission-mode plan option for starting in planning mode. Check the current CLI reference for supported modes and exact behavior, since command-line options can change. Anthropic’s workflow guidance also describes planning before edits so proposed changes can be reviewed before they touch disk.
Implement in small, inspectable steps
Once the plan is accepted, ask Claude Code to make the scoped change rather than expanding the task into unrelated cleanup. Small increments make it easier to spot an incorrect assumption and connect a resulting diff to the intended behavior. For each meaningful step, check that the changed code matches the plan and that the implementation has not silently widened its scope.
At handoff, request a summary of what changed, the commands actually run and their observed outcomes, unresolved risks, and the diff. A claim that tests passed is useful only when the relevant commands were run and their results observed; generated code or a successful completion message is not proof of correctness.
Verify the result in the repository
Verification should follow the change, not merely the plan. Ask Claude Code to run the project’s relevant tests, linters, or other checks, then inspect their output and the final diff. Anthropic’s workflow examples include finding untested code, adding tests and edge cases, running tests, and verifying a refactor.
-
Confirm that checks cover the behavior changed, not just an unrelated fast test.
-
Review failures rather than treating a command’s execution as a pass.
-
Inspect the diff for unintended files, unrelated edits, or missing edge cases.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy. -
If the result differs from the plan, decide whether to revise the code or explicitly revise the plan and checks.
These steps improve inspectability; they do not guarantee that a change is correct. Anthropic’s documentation describes workflows and product features, not an independently measured accuracy or productivity gain for RPI.
Decide whether a subagent is worth the coordination
Delegation is useful when the work is independent and the returned findings justify the extra coordination. Keep a task in the main session when it is small, sequential, centered on one file, or dependent on frequent shared decisions. Anthropic’s prompting guidance recommends subagents for parallel or isolated work while cautioning against excessive delegation for straightforward tasks: Claude Code prompting and best practices.
| Decision factor | A subagent is more useful when… | Stay in the main session when… |
|---|---|---|
| Independence | The subtask can proceed without frequent decisions from the main session. | It depends on continuous shared choices. |
| Context load | Logs, search results, or code excerpts would clutter the main conversation. | The investigation is short and easy to track directly. |
| Parallel value | Distinct investigations can happen at the same time. | The work is inherently sequential. |
| Tools and permissions | The subtask can use a suitably narrow set of tools and permissions. | Separating access would add little practical value. |
| Coordination cost | A concise report is easier to use than doing the investigation in the main thread. | Summarizing and reconciling output would take longer than the task. |
| Usage | The value of delegation justifies its requests against shared usage limits. | The task is straightforward enough to complete directly. |
Choose where subagent definitions belong
Anthropic documents several ways to define subagents: organization-wide managed settings, project-level definitions in .claude/agents/, user-level definitions in ~/.claude/agents/, plugin directories, and agents defined for a CLI session. Project definitions can be checked into version control; user definitions apply at the user level. The documentation describes precedence among locations, so avoid overlapping names unless the intended scope and precedence are clear.
Free tools Windows power users keep installed
One-click scans. No signup required.
The CLI reference also documents --agents for defining session-specific subagents. Definitions include fields for a name, description, prompt, tools, and model. Make the description specific about when the agent should be selected, restrict tools to what the task requires, and define the expected report in its instructions. Because fields, aliases, precedence, and version support may change, consult the live subagent guide and CLI reference before relying on exact configuration details.
A practical RPI checklist
-
Research: Start with the repository’s structure, then trace the relevant behavior, conventions, dependencies, and tests. Delegate only independent or context-heavy exploration.
-
Plan: State the intended behavior, constraints, likely files, risks, and verification steps. Review the proposal before authorizing edits.
-
Implement: Apply the agreed scope in inspectable increments and avoid unrelated changes.
Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy. -
Verify: Run relevant checks, examine their actual results, inspect the diff, and report unresolved risks without claiming checks that were not run.
Quick Recap
SaleBestseller No. 1SaleBestseller No. 2SaleBestseller No. 3Bestseller No. 4
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.
