| Version | Supported |
|---|---|
| 0.1.x | yes |
This is a v0.1 security product. Treat reports against main as in scope.
In scope:
- The preflight scanner executing or importing a scan target
- Zip-slip / symlink escape / path traversal in archive extraction
- Secrets written in full to audit logs, CLI output, or approval reasons
- A DENY decision that a later hook can turn back into ALLOW inside this repository
- Fail-open on high-risk policy / engine errors
- Default upload of source to a remote model
Out of scope:
- Prompt injection that evades the heuristic detector (known limitation)
- Malicious plugins after the user has installed them and granted
allowBuilds/ lifecycle execution - Harness core bugs (report those to deepseek-ai/deepseek-harness)
- Third-party MCP server implementation bugs
Please use GitHub Security Advisories on this repository (Security → Report a vulnerability).
Do not open a public issue for a suspected vulnerability in
dsh-shield itself.
There is no dedicated security mailbox yet. Do not invent one.
Give us a reasonable window to patch before public write-ups. We will credit reporters who want to be named.
- Use synthetic secrets only (
shield_test_…). - Do not commit real credentials, customer data, or live malware.
- Do not execute untrusted plugins to "see what they do."