Security

Controls

Guardrails for SDK and agent workflows.

Generated code, agent actions, and release decisions all cross validation, redaction, and review boundaries before they become trusted output.

Control surface
Security controls stay visible inside the release review loop.
Redacted events

Secrets and credentials are removed before observability events leave the runtime.

Scoped actions

Agent workflows start as dry-runs and declare every planned write before execution.

Typed boundaries

External inputs are validated and converted into explicit artifact contracts.

Release evidence

Every compatibility decision is backed by a report that reviewers can inspect.

Entitlement checks

Expensive actions pass through usage and feature gates before work is dispatched.

Reviewable output

Reports, manifests, and tool results are stable JSON artifacts, not one-off logs.

Readiness checklist
Controls expected before hosted execution.
  • No raw secrets in reports
  • Dry-run first agent execution
  • Typed env validation
  • Auditable run state
  • Least-data observability
  • Organization-ready access model
Evidence packet
Structured output for humans, CI, and agents.
Spec3 operations normalized
Manifest12 public symbols
Compatibilityminor recommendation
Agentdry-run only
Review before write

Agent workflows should produce evidence before they mutate anything.

SDK Parity models agent execution as staged, typed, and inspectable work. Dry-runs, redaction, and audit events are first-class product surfaces.