| name | update-agent-configs |
| description | Research and update secure configuration files for Cursor, Claude Code, Codex, Gemini CLI, Antigravity, and GitHub Copilot CLI. Use when asked to update agent settings or refresh sandbox security practices. |
| license | Apache-2.0 |
| allowed-tools | WebSearch WebFetch |
Update Agent Configs
Purpose
Periodically research the latest security best practices, documentation, and vulnerability reports for AI coding agents (Cursor, Claude Code, Codex, Gemini CLI, Antigravity, GitHub Copilot CLI) and update their respective local configuration files (settings.json, config.toml, sandbox.json, etc.) and markdown guides. This ensures that the security baselines provided by this repository remain effective against emerging supply-chain attacks.
Instructions
- Read References: Read the file
references/documentation-urls.md to get the list of official documentation links for each AI coding agent.
- Fetch Latest Docs & Traverse Links:
- For each tool, use
WebFetch to gather the documentation from the official URLs.
- CRITICAL: Do not just read the main pages. Treat the URLs in
documentation-urls.md as entry points or index pages.
- When reading an index page, identify links to sub-pages (e.g., specific rules, permissions, reference guides, advanced configurations).
- Use
WebFetch to systematically traverse and read these sub-page links to ensure you have a complete and deep understanding of all new settings, hooks, and security properties before moving to the next step.
- Search for Vulnerabilities: Use
WebSearch to search for recent security vulnerabilities related to these agents (e.g., search terms like "AI agent sandbox escape [Current Year]" or "[Agent Name] security vulnerability").
- Compare Configurations: Compare the newly gathered information with the current local configuration files (e.g.,
cursor/sandbox.json, claude-code/settings.json, codex/config.toml, etc.). Look for:
- New settings or flags that should be enabled.
- Deprecated settings that should be removed.
- New default bypasses or exploits that need explicitly mitigating.
- Apply Updates:
- Modify the configuration files to implement the latest secure-by-default practices.
- Update the inline comments in the configuration files to explain the changes.
- Update the respective
README.md files in each agent's directory to reflect the new instructions or threat landscape.
Examples
Example 1: Updating Cursor Settings
Input: "Please update the agent configs."
Execution:
- Agent reads
references/documentation-urls.md.
- Agent uses
WebFetch to read https://cursor.com/docs/agent/security and https://cursor.com/docs/agent/hooks.
- Agent notices sub-page links related to advanced sandboxing properties in the fetched content.
- Agent uses
WebFetch to follow those links and thoroughly reads the nested sandbox reference.
- Agent discovers a new property
restrictSubprocesses added to sandbox.json in a recent update.
- Agent updates
cursor/sandbox.json to include "restrictSubprocesses": true.
- Agent updates
cursor/README.md to document the new property and how it mitigates attacks.
Additional Resources