@mode-db/server@0.1.12

High-risk behavior combination matched malicious policy.

npm install of @mode-db/server@0.1.12

AI assistants operating in the project may ingest package-supplied workspace instructions without explicit user action.

postinstall copies package-supplied AGENTS.md into consumer workspace

On install, npm runs the package postinstall script. The script resolves the consumer workspace from INIT_CWD or process.cwd, creates a modeldb directory, and copies the package's AGENTS.md into that workspace if absent. That file is an AI assistant instruction surface, so the package can influence future agent behavior without the user invoking the CLI or opting into workspace instruction installation.

Static source inspection confirms lifecycle-time creation of a package-supplied AGENTS.md in the consuming project. Under the firewall policy this is unconsented AI-agent control-surface mutation and should be blocked despite otherwise package-aligned server code. Product guard normalized a concrete AI-agent control hijack publish_block to the blockable dangerous-capability shape.