Updated 2026-07-24

Agent rules snippets

Paste these snippets into your agent rules file to teach your AI coding assistant when and how to invoke LyraShield checks automatically. Adapt the section headings to match your project's existing conventions.

CLAUDE.md (Claude Code)

Add this section to your CLAUDE.md file in the project root. Claude Code reads this file at the start of every session.

## LyraShield security checks

LyraShield is available as an MCP tool. Use it to verify security posture before
creating a pull request and after applying a security fix.

### Before creating a PR
Run `check_diff` on the staged changes to identify security issues introduced
by this work item:
1. Use the `lyrashield` MCP tool `check_diff` with the diff of staged changes.
2. Review any findings before committing.
3. If findings are reported, address them or document why each is acceptable.

### After applying a security patch
Run `verify_fix` to confirm that the fix resolves the reported condition:
1. Apply the fix.
2. Use the `lyrashield` MCP tool `verify_fix` with the finding ID.
3. Include the verification result in the PR description.

### When not to run checks
- Do not run checks on files you do not have permission to scan.
- Do not run scans against third-party URLs without explicit authorization.
- A clean check result does not guarantee the absence of all vulnerabilities.

AGENTS.md (JetBrains Junie, OpenAI Codex, Hermes)

AGENTS.md is read by JetBrains Junie, OpenAI Codex, and other agents that follow the OpenAI Agents spec. Place it in your project root.

## LyraShield security workflow

The LyraShield MCP server is available as a registered tool.

### Pre-PR check
Before opening a pull request, run check_diff on the changes in this branch.
Report any findings and wait for resolution or explicit acceptance before proceeding.

### Post-fix verification
After applying a fix for a security finding, run verify_fix with the finding ID.
Include the verification receipt in the PR description.

### Scope limits
Only run security checks against targets that are owned by this workspace and
explicitly listed as authorized targets in the LyraShield settings.

.cursorrules (Cursor)

Cursor reads .cursorrules from the project root for project-specific AI behavior. Add this block to your existing file or create a new one.

# LyraShield security checks

When preparing a pull request:
- Use the lyrashield MCP check_diff tool to scan staged changes for security issues.
- Do not create the PR until findings are reviewed.

When fixing a reported security issue:
- After applying the fix, use lyrashield verify_fix with the finding ID.
- Include the verification result in the PR description.

Constraints:
- Only scan authorized targets listed in the LyraShield workspace.
- A passing check is not a guarantee of zero vulnerabilities.

Tips for effective agent rules

• Keep the rules focused on when to run checks — the agent will use the MCP tool descriptions to understand how.
• Add an explicit scope constraint to prevent the agent from scanning targets it does not own.
• Include a note that a passing check is not a guarantee — this keeps expectations accurate and prevents over-reliance on automated results.
• For teams, commit the rules file to the repository so all developers share the same behavior.