Align tool annotations with ChatGPT plugin guidelines

[rgarcia]

Summary

The ChatGPT plugin review flagged tool annotations that don't match tool behavior. This changes five hints to follow the published rules (read-only only for pure retrieval; destructive for any delete, overwrite, cancel or stop; open world for public or arbitrary destinations):

tool	hint	before → after	why
manage_config_registry	destructiveHint	false → true	cancel_analysis cancels a running analysis
manage_replays	destructiveHint	false → true	stop ends a recording irreversibly
get_more_tools	openWorldHint	true → false	only files a report with KERNEL
manage_proxies	openWorldHint	false → true	check and custom proxies reach arbitrary hosts
web_search	readOnlyHint	false → true	search and content retrieval don't modify state; the guidelines name web search as read-only

Adds a test in register.test.ts that pins readOnlyHint, destructiveHint and openWorldHint for every registered tool, so each hint is an explicit boolean and any future change is deliberate.

Testing

• bun test: 888 pass, 0 fail
• tsc --noEmit clean

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
mcp	Ready	Preview	Oct 5, 2026 7:00pm UTC