Engineering simulator

Simulate a governed AI coding-agent run before trusting live execution.

This static run path helps a team evaluate fit before live access: bind a requirement, map the software workflow, attach verification records, and define the evidence boundary for human approval.

Start from this demo Request API access View workflow spec Open review JSON

SDLC WorkGraph plan

Fix failing web build after dependency update

6 steps · 6 lanes · safe-stop: Discard generated handoff or patch draft

Bind repo objective and approval boundary
Inspect public repo metadata and failure context read-only
Draft bounded patch or explicit blocker route
Run available lint/typecheck/test/build verification receipts
Capture risks, safe-stop path, and human approval request
Emit Evidence Report handoff

01planner

Bind repo objective and approval boundary

passed

02repo-inspector

Inspect public repo metadata and failure context read-only

passed

03code

Draft bounded patch or explicit blocker route

passed

04terminal

Run available lint/typecheck/test/build verification receipts

passed

05quality

Capture risks, safe-stop path, and human approval request

waiting for human

06review

Emit Evidence Report handoff

waiting for human

Lint

Code/style sanity

repo CI conventions

Typecheck

Contract correctness

TypeScript/Python schema gates

Unit tests

Behavior regression proof

GitClaw/GitWork test discipline

Build

Deploy readiness

Vercel/App Router gates

Policy checks

Governance and approval boundary

GitClaw WorkGraph policy

Security scan

Risk screening

DeepBrainz guardrail taxonomy

Regression/eval

Quality signal beyond tests

DeepBrainz eval vendors: lm-evaluation-harness, lighteval, simple-evals, verifiers, frontier-evals

Human review

Approval authority

Human governance gate

Edit inputs, then draft an Evidence Report · sample review JSON