Skip to content

fix: declare explicit annotation hints on send_feedback - #57

Merged
fatwang2 merged 1 commit into
mainfrom
claude/project-thread-cjr0p2
Oct 31, 2026
Merged

fatwang2 merged 1 commit into
mainfrom
claude/project-thread-cjr0p2

Conversation

@fatwang2

Copy link
Copy Markdown
Member

Requested by fatwang2 · project thread

Why

ChatGPT's plugin guidelines (developers.openai.com/plugins/plugin-guidelines, "Correct annotation") require readOnlyHint, destructiveHint and openWorldHint as explicit booleans on every tool. OpenAI scans our published plugin's MCP server daily and holds new tools that fail checks. send_feedback comes from the Agentback SDK (agentback@0.2.0), which registers it with only readOnlyHint: false, openWorldHint: true, so destructiveHint is missing.

What

  • After attachToMcp, update the registered tool's annotations to readOnlyHint: false, destructiveHint: false (a report only adds a record), idempotentHint: false, openWorldHint: true.
  • Test asserts the served send_feedback annotations.
  • CHANGELOG Unreleased entry.

The longer-term fix is to set destructiveHint: false in the Agentback SDK itself; this keeps search1api-mcp correct regardless of SDK version.

Testing

npm test: 51/51 pass.


Generated by Claude Code

The Agentback SDK registers send_feedback without destructiveHint.
ChatGPT's plugin guidelines require readOnlyHint, destructiveHint and
openWorldHint as explicit booleans on every tool, and the daily MCP
scan holds tools that miss them. Set all hints after registration.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DuYf7tfmsW91uukAP5riQ4
@fatwang2 fatwang2 self-assigned this Oct 31, 2026
@fatwang2
fatwang2 merged commit 9ea82a7 into main Oct 31, 2026
3 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants