Skip to content

Security: generalbusiness-ai/dap

Security

SECURITY.md

Security

Report a vulnerability

Use GitHub private vulnerability reporting for dap. Include the affected commit or deployment, the request or command, the expected boundary, and a minimal reproduction. Omit signing keys, bearer secrets and private room content. If GitHub offers no private reporting form, contact the repository owner through GitHub to arrange a private channel before sharing the reproduction.

Security references

Read the references in this order. The first three explain the model. The next two are inventories for lookup. The last item holds the normative contracts.

  1. Threat model: what dap protects, what an attacker can do, the invariants and the tests that check them.
  2. Deployment trust boundaries: secret custody, shared resources and revocation limits.
  3. Definition-author security reference: validation refusals and valid permissive policy forms.
  4. Worker entry points and gates: routes, caller gates, pre-gate work, stored state, bounds and outside calls.
  5. Worker settings and bindings: setting readers, parsers, failure modes and health meanings.
  6. The normative contracts: the scope and replay contract and authority, effects and publication. Their section numbers are stable.

Testing guidance describes the evidence required for a change. The inventories describe checked-in source. They establish no current deployed setting, provider access control or completed live scene.

Reviews

The 4 October security review covers the earlier model at its recorded commit. It remains historical evidence and does not assess the current implementation. The 5 October dispositions record the treatment of that review period's findings.

The attack-surface review of 21 October 2026 records its findings at commit 603220b9. The repository's gitseq workroom holds that review, its remediation requests and the later review of each fix. The delivery note for the settings, trust-boundary and definition-author pages and the completion note for the threat model name the documentation evidence. A review of one commit proves no later commit or deployed configuration.

There aren't any published security advisories