Learnings & Knowledge Base

Teach CodePeel your codebase conventions. Learnings are rules that get injected into every review, so the AI checks for your team's specific patterns.

Adding a Learning

Via PR Comment (fastest)

Mention @codepeel with the learn: prefix in any PR comment:

@codepeel learn: Always use Zod schemas for API request validation

@codepeel learn: We use effect-ts for error handling, not try/catch

@codepeel learn: The src/generated/ directory is auto-generated — never flag issues there

The rule is saved immediately and applied to all future reviews for this repository.

Via .codepeel.yml (version-controlled)

Add rules to your config file for team-wide conventions:

expert_rules:
  - "Use functional components with hooks, not class components"
  - "All database queries must use the repository pattern"
  - "API responses follow JSend specification"

The difference: learn: rules are stored in CodePeel's database per-repo. expert_rules live in your repo and are version-controlled.

---

Adding Ignore Rules

Suppress false positives with the ignore: prefix:

@codepeel ignore: Don't flag TODO comments as issues

@codepeel ignore: The console.log in src/lib/debug.ts is intentional

---

Viewing Your Learnings in the Webapp

Knowledge Base Page

Go to Knowledge Base in the sidebar. This shows:

• Learning growth over time (chart)
• Pattern category breakdown (pie chart)
• Learnings per repository
• Recent learnings list

Use the date range filter (7d / 30d / 90d) or filter by repository.

Learnings Page

Go to Learnings in the sidebar. This shows the per-repository list:

1. Select a repository from the dropdown
2. View all active learnings for that repo
3. Use thumbs up/down to give feedback
4. Delete learnings you no longer need

---

Managing Learnings (Thumbs Up / Down)

Each learning has feedback buttons:

---

How Learnings Affect Reviews

When CodePeel reviews a PR:

1. Loads all learnings for the repository
2. Injects confirmed rules as "TEAM PREFERENCES" in the AI prompt
3. Injects thumbs-down rules as "FALSE POSITIVE CORRECTIONS" (don't flag these)
4. The AI considers these alongside its standard analysis

The more specific your learnings, the more relevant your reviews become.

---

Best Practices

• Add learnings early — the sooner CodePeel knows your conventions, the better
• Be specific — "Use parameterized queries" is better than "Be careful with databases"
• Review periodically — remove outdated rules from the Learnings page
• Use ignore rules on the spot — when you see a false positive, suppress it immediately
• Combine both methods — expert_rules for team standards, learn: for ad-hoc rules