SDLC Workflows With The Autohand SDK
May 5, 2026 ยท View on GitHub
These workflows use the SDK as an inspectable orchestration layer around the Autohand CLI. The common pattern is:
- Start in plan mode for read-only discovery.
- Review the generated plan outside the agent loop.
- Disable plan mode only after approval.
- Execute with explicit permission handling.
- Run release gates and summarize residual risk.
Discovery And Planning
Use 20-sdlc-discovery-plan.ts when the task is still ambiguous. It starts the CLI in plan mode with:
const sdk = new AutohandSDK({
planMode: true,
skills: ['typescript', 'testing'],
});
Plan mode is separate from permissionMode. It calls CLI-3
autohand.planModeSet and restricts the agent to read-only planning tools.
Gated Implementation
Use 21-sdlc-gated-implementation.ts when you want a two-phase workflow:
- Generate a plan with
planMode: true. - Stop by default.
- Re-run with
AUTOHAND_EXECUTE_PLAN=1to disable plan mode and implement.
This is useful in CI-like hosts or IDE integrations where the host owns the approval gate.
Release Readiness
Use 22-sdlc-release-readiness.ts to ask the agent to run:
bun run typecheck
bun run test
bun run build
bun run lint
Keep permissionMode: 'interactive' so command execution remains visible through
permission_request events.
Plan Mode Support
The SDK now supports plan mode directly:
const sdk = new AutohandSDK({ planMode: true });
await sdk.start();
await sdk.disablePlanMode();
await sdk.enablePlanMode();
await sdk.setPlanMode(false);
permissionMode: 'plan' remains accepted as a legacy configuration value, but
new code should prefer planMode and setPlanMode() because CLI-3 exposes plan
mode as its own RPC control.