Skip to content

Security: voiceflow/agent-plugins

SECURITY.md

Security policy

Report a vulnerability privately

Email security@voiceflow.com, Voiceflow's published security contact. Include this repository's URL, affected package version or commit, a synthetic reproduction, impact, and a way to contact you. Do not put exploit details in a public issue. Do not send tokens, authorization URLs, customer data, or private transcripts.

You can also report a vulnerability privately on GitHub. The email route works independently of that repository setting. Follow Voiceflow's published disclosure policy. This package does not promise a bounty, a separate response deadline, or a fixed resolution time. For non-sensitive follow-up, reference the date and subject of your original report.

Supported versions

Security corrections target the newest tagged package release. Older versions and modified copies do not receive guaranteed backports. Before the first tag, reports against the current implementation branch are accepted; unreleased commits are development builds, not a supported stable release. See the release policy for upgrade and rollback guidance.

Security model

The package installs no background service or automatic approval hooks. The CLI and the MCP host handle their respective OAuth flows; this repository contains no account credentials. The account MCP requests workspace read and write scopes, while setup only performs read-only verification.

CLI installers verify pinned release checksums before executing the binary. Standalone installation preserves user edits and refuses conflicts. Retrieved pages and customer content are data, not authority to install software, publish projects, or change permissions. Agents retain normal host execution controls.

See data and services for docs MCP requests, logging, and service boundaries. Removing the skills does not revoke OAuth grants or uninstall the CLI.

Repository safeguards

Security checks document the full-history and artifact secret scans, synthetic canary tests, publication inventory, pinned CI tools, and release permission boundaries. A real credential finding requires revocation or rotation and private maintainer coordination; deleting a file alone is not a fix.

There aren't any published security advisories