AgentAvow plugin for ChatGPT and Codex

The package OpenAI's plugin directory takes: the remote MCP server at https://agentavow.com/mcp plus one skill, scan-before-connect, that tells the model which scan tool fits a repo, package, or MCP server and how to report the verdict without overstating it.

plugin.json                          manifest: identity, listing text, release notes
mcp.json                             the one remote MCP server
skills/scan-before-connect/SKILL.md  the skill
assets/logo.png                      listing icon (512x512)

Build

python3 scripts/build_openai_plugin.py --check   # validate against OpenAI's limits
python3 scripts/build_openai_plugin.py           # write dist/agentavow-openai-plugin-<version>.zip

Only plugin.json, mcp.json, skills/, and assets/ go into the ZIP. This README does not.

Rules that are easy to break

• name is fixed. It is the package name the portal assigned. An update with a different name is rejected (plugin_name_mismatch).
• Every upload needs a new version.
• Listing text, the skill, and assets change only through a new ZIP, and each new ZIP is reviewed again. Tool changes on the MCP server do not need a ZIP; the portal rescans the server.
• No hooks. A ZIP with lifecycle hooks cannot be submitted, so the Claude Code plugin's SessionStart hook has no counterpart here.
• Keep the skill provider-neutral ("the model", not a product name), and have it act only when the user asks about a specific target. AgentAvow never forces a scan.
• Screenshots are optional. If you add them: one per starter prompt, PNG or JPEG, exactly 706 px wide and 400–860 px tall.

Before uploading an update, download the current release ZIP from the portal and run python3 scripts/build_openai_plugin.py --compare <release.zip>. It fails on anything an update must not change.