What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Configure a coding agent for a pre-edit investigation: ask it one specific question about the existing repository, require a source-backed map of relevant files and tests, and verify its findings before asking it to implement anything. Keep durable project rules in the instruction files recognized by your agent’s harness, then check that those files are being discovered and applied.
How to make an agent inspect a repository before coding
- Choose one behavior to trace. Ask a concrete question about the change you plan to make—for example, where a request is authorized, where a form saves data, or where an API response is assembled. Avoid broad prompts such as “understand the whole codebase.” A bounded question focuses the search on useful evidence. Visual Studio Code’s codebase-exploration guide uses this kind of focused investigation.
- Make investigation a separate first phase. Tell the agent not to edit files or generate implementation code yet. Ask it to identify likely entry points, follow the relevant calls, locate associated tests, and report any unresolved questions. Request a concise list of files with supporting source references—not a speculative overview of the repository.
- Review the map against the source. Treat the agent’s explanation as a hypothesis, then open the cited files and confirm that the code supports its account. Source reading can begin without installing dependencies or running the application; consult the project’s setup instructions before deciding whether runtime confirmation is necessary. VS Code’s guide describes this source-first approach.
- Pass confirmed context into the implementation request. Once the relevant files are clear, give the agent those files and the investigation findings as context for the next task. This avoids asking it to repeat a broad repository search and lets you correct mistaken assumptions before code changes begin.
A reusable investigation prompt
Adapt this prompt to the behavior you need to change:
Before editing anything, investigate how [specific behavior] works in this repository. Identify the likely entry point, trace the relevant calls and data flow, and find related tests. Return a concise map of the relevant files with source references, explain what each file contributes, and list unresolved questions. Do not modify files or propose implementation code yet.
After reviewing the report, decide whether the findings are sufficient to proceed. If a claim is unclear or unsupported, ask the agent to inspect the specific source or test rather than expanding the task into a repository-wide summary.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →#1 Best Overall
Where to put persistent coding-agent instructions
Use the instruction-file name and location documented for the harness you actually use. A filename that works in one editor or agent may be ignored by another. VS Code’s codebase-customization guide recommends AGENTS.md for OpenAI Codex, .github/copilot-instructions.md for GitHub Copilot, and CLAUDE.md for Claude Code.
| Harness or use | Documented instruction mechanism | Scope |
|---|---|---|
| OpenAI Codex in the VS Code guide | AGENTS.md (VS Code customization guide) |
Repository guidance; check the harness’s discovery behavior for your setup. |
| GitHub Copilot repository-wide instructions | .github/copilot-instructions.md (VS Code customization guide; GitHub code review documentation) |
Repository-wide review or coding context. |
| GitHub Copilot path-specific instructions | .github/instructions/**/*.instructions.md (VS Code custom-instructions documentation) |
Instructions for matching paths. |
| Claude Code | CLAUDE.md or files in .claude/rules (VS Code customization guide; VS Code custom-instructions documentation) |
Project guidance, with path-specific rules available through the documented rules mechanism. |
Names and discovery rules can change, and configuration support depends on the harness. Check its current documentation rather than assuming that a file listed for one tool will work in another.
What belongs in an instruction file
Use persistent instructions for expectations that matter across tasks and are not reliably inferable from the source—for example, a required verification approach or a project decision that the code does not make clear. Keep instructions focused. Duplicating obvious facts adds text without making the agent’s investigation more reliable.
Use repository-wide instructions for conventions that apply broadly. Where the harness supports them, use path-specific instructions for local rules that apply only to matching files. VS Code documents .github/instructions/**/*.instructions.md for Copilot and .claude/rules for Claude; the exact matching and loading behavior belongs to each harness. See its custom-instructions guide.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Clear out junk files and repair common Windows errors3Fix the driver behind crashes, sound loss and screen glitchesHow to check whether instructions and modes are working
- Confirm the mechanism: verify the expected filename, directory, and harness against current product documentation.
- Check scope: determine whether the rule is repository-wide or applies only to files matching a path pattern.
- Test the pre-edit request: give the agent a bounded investigation task and check whether it returns file references, a behavior path, tests, and unresolved questions before making changes.
- Inspect loaded instructions where the product allows it: use the harness’s available instruction or context view rather than assuming that a file was read.
- Correct configuration before adding more text: if a rule seems ignored, investigate its name, location, harness support, and scope first.
VS Code’s documentation covers instruction discovery and scope across its supported customization mechanisms. Consult the custom-instructions guide for those mechanisms.
Use a read-only mode only when the product documents one
Some tools offer modes that separate searching and answering from editing, but mode names and guarantees are product-specific. Cursor’s modes documentation describes Ask as a way to search a codebase and answer without making changes, and Manual as a mode for explicitly selected file edits without searching or running commands. That distinction can help when choosing an investigation workflow, but it should not be assumed to describe other agents—or to replace checking what the selected mode actually permits. See Cursor’s modes documentation.
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.
