Skip to content

Security: langchain-ai/paid-media-agent

Security

SECURITY.md

Security policy

Reporting a vulnerability

Use GitHub's private vulnerability reporting on this repository ("Report a vulnerability" under the Security tab). Do not open a public issue for a security problem and do not include account identifiers, tokens, or provider payloads in a report.

What counts

  • Any path that lets model output, Slack input, MCP metadata, or a file reach a provider mutation without a host-created, signed, single-use approval.
  • Any way to reach a mutation tool, a raw mutate endpoint, or a delete from the model's tool surface.
  • Approval replay, edit-without-reapproval, digest bypass, expired or foreign approvals, or a proposal executing on a thread other than its own.
  • Credential exposure in prompts, tool results, artifacts, logs, Slack blocks, or receipts.
  • Path traversal out of the workspace, artifact delivery of unexpected files, or Slack signature verification bypass.

What we commit to

  • Acknowledge reports within five business days.
  • Keep the fixture-only default: no live provider mutation is reachable without the operator gates in docs/operations/live-write-runbook.md.
  • Publish fixes with a changelog entry and, where relevant, a regression test in tests/.

Supported versions

Only the main branch is supported before the first tagged release.

There aren't any published security advisories