Trust center / security
Trust boundaries are enforced below the interface.
Web, API, CLI, and MCP calls converge on organization-scoped services. Transport convenience never replaces authorization or ownership checks.
Identity and tenant isolation
Credentials are hashed or cryptographically verified, scopes are explicit, and repository queries require the authenticated organization. Caller-supplied tenant identity is not trusted.
Source acquisition
URL adapters accept public HTTPS sources within provider and product policy. Redirect targets, addresses, content limits, and acquisition failures are checked before processing.
Files and artifacts
Uploads use short-lived intents, exact content metadata, SHA-256 checksums, scanning states, and organization ownership. Output manifests carry checksums and signed access expires.
Agent boundary
Hosted MCP receives only the OAuth public verification key and never reads host paths. Local MCP resolves paths inside explicit allowed roots and keeps protocol stdout separate from diagnostics.
Reporting
A public vulnerability-reporting address and response policy must be configured before general availability. Until then, use the repository security contact supplied with your deployment.