Free tools Windows power users keep installed
One-click scans. No signup required.
To stop an AI coding assistant from repeating a mistake, turn the correction into a short, testable instruction and save it where the assistant will load it in future sessions. Put it in the narrowest useful scope, then open a fresh session and confirm it was loaded. Saved instructions improve continuity, but they are guidance—not a guarantee. For actions that must be blocked, use an available client-enforced permission, hook, or policy control.
Why does the mistake come back in a new session?
A new session may start without the previous conversation. Claude Code’s documentation puts it plainly: “Each Claude Code session begins with a fresh context window.” If the correction exists only in chat history, a later session may not see it. Persistent instruction files and memory features can carry selected guidance forward, but their names, scopes, and loading rules vary by product and session mode.
As an Amazon Associate I earn from qualifying purchases.
The useful goal is not to preserve the whole conversation. It is to save the project fact or behavioral rule that would have prevented the mistake.
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 →How do you turn a correction into a useful rule?
Write down what triggered the mistake, what the assistant should do instead, and how someone can check the result. Prefer a behavior a reviewer can observe over a broad request such as “write clean code.”
| Weak correction | More actionable instruction |
|---|---|
| “Don’t break the tests.” | “Before changing the parser, run the parser test suite. Add or update a test for each changed parsing behavior, and report the command and result.” |
| “Use our conventions.” | “For files under src/api/, use the existing request-validation helper. Do not add a second validation library.” |
| “You forgot the migration again.” | “When changing a database schema, include a migration and verify it using the project’s documented migration check.” |
These examples are templates, not claims about a particular project. Replace the paths and checks with conventions and commands that are actually valid in your repository.
Where should the instruction live?
Choose the smallest scope that will reliably reach the work it governs. A team convention usually belongs in the repository’s supported instruction file; a personal preference belongs in personal settings or memory; a rule for only certain files belongs in path-specific instructions when the assistant supports them. A one-off procedure is better kept in an on-demand mechanism than loaded into every task.
| Scope or mechanism | Use it for | Watch out for |
|---|---|---|
| Personal instructions or memory | Your preferences or recurring guidance that should follow you across projects. | It may not be shared with teammates or apply to every assistant and mode. |
| Repository instructions | Shared conventions, architecture facts, and build or validation steps for a project. | Keep the always-loaded file focused; an irrelevant rule can distract from the task. |
| Path-scoped rules | Conventions that apply only to specified files or directories. | Support and discovery behavior vary by product. |
| On-demand procedures | Long or occasional workflows that should not occupy every session’s context. | The assistant may need to be told when to load or use them. |
| Client-enforced controls | Actions that must be restricted independently of the assistant’s judgment. | Availability and configuration are product-specific; instructions alone are not enforcement. |
How do you set this up in common coding assistants?
Claude Code
Anthropic documents human-maintained CLAUDE.md instructions separately from auto memory, which stores assistant-written learnings and patterns. It also documents multiple instruction scopes and rules in .claude/rules/, including path-specific rules. Auto memory loads up to the first 200 lines or 25KB by default, so keep important always-relevant guidance concise. Claude Code documents /context for checking loaded files and /doctor prompt-audit for supported versions. Availability of commands and features can depend on the installed version and settings.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →Claude Code also documents optional AGENTS.md support with version and configuration qualifications. Do not assume that file is being read: verify support in your installed setup.
GitHub Copilot
GitHub documents repository-wide custom instructions, path-specific instruction files, and AGENTS.md agent instructions, along with controls for reviewing and managing stored Copilot memories. Its onboarding guidance recommends covering the repository’s purpose, layout, architecture, build steps, and validation steps, and keeping instructions task-independent. GitHub’s guidance proposes a two-page maximum for generated repository instructions; treat that as vendor-specific advice, not a universal size limit.
VS Code
Visual Studio Code documents reusable Markdown custom instructions and multiple file types. Whether an instruction file is discovered depends on the selected agent harness, such as the Local agent or an Agent Host harness. Check which agent is active rather than assuming that a file supported in one mode will load in another.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How can you verify the fix in a fresh session?
- Save the instruction in the product’s supported file, memory, or settings location, using the intended project or path scope.
- Start a new session in the same assistant and mode where you expect the rule to apply.
- Check loaded context using the product’s available context or memory view. For Claude Code, the documented
/contextcommand can show loaded files. - Give the assistant a relevant task and inspect whether it follows the rule. Check the outcome against the rule’s stated test or validation step.
- Review instructions periodically and remove stale or contradictory guidance. Overlapping rules can make behavior unpredictable.
If the rule does not appear in the loaded context, troubleshoot discovery and scope before rewriting it. Confirm the filename, location, selected mode, and product version against current documentation.
When are instructions not enough?
Persistent instructions shape model behavior; they do not guarantee compliance. If a mistake could cause a prohibited or damaging action, use a product-supported permission setting, hook, or policy control where available. Keep the instruction as useful context, but make the client enforce the restriction rather than relying only on a reminder in text.
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.
