| Version | Supported |
|---|---|
| 0.1.x | Yes (current) |
| < 0.1.0 | No |
Please do not report security vulnerabilities through public GitHub issues.
Use GitHub's private vulnerability reporting to submit a security advisory directly to the maintainers. This keeps the report private until a fix is available.
Email runzaisongpu95@gmail.com with subject [tab-conductor SECURITY] <brief description>.
Include:
- Description of the vulnerability
- Steps to reproduce
- Affected version(s)
- Potential impact assessment
- (Optional) Suggested fix
tab-conductor follows a 90-day coordinated disclosure policy:
- You report the vulnerability privately.
- Maintainers acknowledge receipt within 5 business days.
- Maintainers investigate and develop a fix.
- A patched release is published within 90 days of the initial report.
- After the patch is available (or 90 days, whichever comes first), you are free to disclose publicly.
tab-conductor implements a 4-layer security model:
| Layer | Countermeasure |
|---|---|
| Prompt level | --append-system-prompt instructs workers not to output secrets or override supervisor |
| Tool level | --disallowedTools baseline blocks sudo *, curl * | sh, rm -rf /*, and credential file reads |
| State level | JSON Schema 2020-12 strict validation; optional HMAC-SHA256 signing (hmac_signing: true) |
| Network/Env level | SecretFilter regex scrub before any write; filtered environment inheritance; bugreport redaction |
Full threat model: docs/SECURITY_THREAT_MODEL.md
The following are in scope for security reports:
- Secret leakage through state.json, log files, or bugreport packages
- State tampering enabling cost cap bypass or unauthorized worker spawning
- Prompt injection enabling workers to read or exfiltrate credential files
- Path traversal in
secret_filter.pyorstate.pysymlink handling - HMAC bypass when
hmac_signing: true
The following are out of scope:
- Vulnerabilities in the
claudeCLI itself (report to Anthropic) - Vulnerabilities in the operating system or Python interpreter
- Social engineering attacks