> ## Documentation Index
> Fetch the complete documentation index at: https://doc.entelligence.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Guidelines, Learnings, and codebase context
> Teach Code Review your architecture and standards with guidelines, Learnings, Markdown, linked codebases, production evidence, and exclusions.
Teach Code Review what good looks like in your repositories. Stable guidelines define the standard; Learnings capture reviewed exceptions; repository and production context explain how the system actually behaves.
## Choose the right context source
| Context source | Best for |
| ------------------------------- | ------------------------------------------------------------------------------------------- |
| Guidelines | Architecture rules, language practices, security standards, and review policy |
| Learnings | A recurring false positive or a repository-specific exception confirmed by the team |
| Repository Markdown | Architecture notes, contribution guidance, AGENTS.md, and instructions stored with the code |
| Linked codebases | Service-to-SDK, schema-to-consumer, or shared-library relationships |
| Production and incident context | Patterns from connected alerts, regressions, and reliability evidence |
| File exclusions | Generated or low-value paths that should not consume review attention |
## Write guidelines that produce useful findings
In **Code Review → Codebases → Guidelines**, upload a guideline or generate one from repository review history, then assign it to the intended codebases.
A strong guideline:
* Defines one observable behavior.
* Names the repository, language, or path scope.
* Explains why the rule matters.
* Includes a short example when the rule could be interpreted more than one way.
* Has an owner who reviews it as the architecture changes.
Avoid duplicating generic formatter or linter rules unless the reviewer needs the underlying engineering context.
## Turn feedback into Learnings
When a reviewer marks a finding as unhelpful and explains why, Entelligence can use that feedback to reduce the same false positive in later reviews.
Review the **Learnings** tab periodically:
* Keep confirmed repository exceptions.
* Narrow Learnings that suppress more than the intended pattern.
* Remove entries that are obsolete after a refactor or policy change.
* Fix real problems in code or policy instead of teaching the reviewer to ignore them.
## Use repository Markdown
Enable maintained Markdown files from **Codebase Context**. Good sources include architecture decisions, contribution guidance, local agent instructions, and operational notes.
Stale instructions create stale reviews. Give each context file an owner and update it with the codebase.
## Connect related systems
Link repositories when a change in one can break another—for example, an API and its SDK, a schema and its consumers, or a shared library and downstream applications.
When production or incident integrations are connected, relevant operational evidence can help the reviewer recognize recurring failure patterns. Keep these connections within approved workspace and repository boundaries.
## Exclude files deliberately
Built-in groups cover migrations, snapshots, generated code, and test fixtures. Custom glob patterns handle repository-specific paths.
Before enabling an exclusion:
1. Identify every path it matches.
2. Use the narrowest pattern.
3. Review one pull request containing both an excluded and included file.
4. Record the owner and reason.
5. Revisit the exclusion when build or generation behavior changes.
Exclusions reduce noise by reducing coverage. Never exclude a production-affecting file only to improve a score.
## Verify the context
Create a small test change that should trigger one team rule and one built-in check. Confirm that the review:
* Cites the assigned guideline.
* Ignores disabled Markdown context.
* Applies the intended Learning.
* Checks linked repositories only where configured.
* Uses production evidence only from approved integrations.
* Skips exactly the excluded paths.
## Next step
Tune [Code Review settings](/code-review/settings) or resolve [context and permission issues](/code-review/troubleshooting).