| Version | Supported |
|---|---|
0.1.x |
Yes |
< 0.1 |
No |
If you discover a security vulnerability in W Agent (this repository’s code, default configs, or documented deploy paths), please report it privately so we can fix it before public detail is useful to attackers.
Please do
- Email or open a GitHub Security Advisory (preferred when available on the repo).
- Include: affected version/commit, component (
src/safety, MCP HTTP, dashboard auth, bridge, etc.), reproduction steps, and impact. - Allow a reasonable window for a fix before public disclosure (we aim to acknowledge within 72 hours and ship or publish guidance as soon as practical).
Please do not
- Open a public GitHub issue with exploit detail for unpatched flaws
- Access data or accounts that are not yours while testing
- Run injection or spam campaigns against third-party WhatsApp numbers
| Report | Why |
|---|---|
| WhatsApp / Meta account bans from Baileys or similar unofficial clients | Platform policy and detection — see docs/security.md and docs/providers.md. Not a patchable application CVE. |
Issues that require a stolen .env, DASHBOARD_TOKEN, MCP_AUTH_TOKEN, or Baileys auth directory |
Expected failure mode if secrets are exposed |
Prompt-injection “bypasses” that only work when the operator pastes attacker text as a trusted owner /agent command |
Owner channel is intentionally trusted |
| Denial of service against your own self-hosted instance | Operator environment |
Operators should read docs/security.md for the full threat model. In short:
- Inbound WhatsApp bodies are treated as untrusted data (delimiters + classifier; silent mode on high score)
- Outbound traffic defaults to draft → owner approve (MCP included)
- Auto-sends are rate-limited (quotas, quiet hours, daily cap)
- Remote MCP requires
MCP_AUTH_TOKEN; dashboard requiresDASHBOARD_TOKEN - Chat history is stored in your Postgres; protect backups and auth volumes
pnpm redteam # canned injection payloads against the local guard (no network send)Do not use production personal WhatsApp accounts for bridge or red-team experiments.
- Security advisories: GitHub Security Advisories on Chessing234/w-agent
- Non-security bugs: GitHub Issues