AI Decision System
Replayable, offline-verifiable records for AI-assisted go/no-go decisions: experiment promotions, typed model answers and human-approved actions.
Design, plan, create, and validate supported apps with Autograph App Builder.
Developer toolsTypeScript · JavaScriptGitHub @withAutographOnlineOwner unconfirmed
Have you tried it? One tap is enough. No sign-in needed.
Autograph App Builder Eve agent and Codex plugin
Taken from the repository README
Loading the full README. The saved excerpt is shown below.
Autograph App Builder is a portable Agent Plugin
for turning product ideas into working applications. Codex is the first
entrypoint, and the public MCP contract exposes exactly five tools:
autograph_start, autograph_get, autograph_send, autograph_respond, and
autograph_cancel.
Complete OAuth, open a new task, and mention @Autograph App Builder. See
installing for client-specific details.Read the full README ↗Autograph App Builder is a portable Agent Plugin
for turning product ideas into working applications. Codex is the first
entrypoint, and the public MCP contract exposes exactly five tools:
autograph_start, autograph_get, autograph_send, autograph_respond, and
autograph_cancel.
codex plugin marketplace add withAutograph/marketplace
codex plugin add app-builder@autograph
Complete OAuth, open a new task, and mention @Autograph App Builder. See
installing for client-specific details.
Use mise run dev. Edit live code, let HMR or the affected process reload, and
retry the behavior. Vercel Sandbox is the execution backend and project-scoped
OIDC is the provider credential boundary. The checkout is writable and new or
generated files are expected.
Do not rebuild packages, run broad suites, publish, or deploy after every edit. Caches are optional accelerators, and inspection is best-effort. The builder does not block on exact versions, source drift, package layout, topology, receipts, digests, quotas, or other speculative conditions. See the local-development lifecycle and execution boundaries.
Use mise run eval:design -- --help to score an existing preview on demand.
Design-quality evals report Arrusted token
evidence, responsive/browser findings, and an advisory AI design score without
regenerating the app or adding a runtime gate.
Use --additional-desktop-size WIDTHxHEIGHT to add one explicit desktop window
observation; the default three desktop captures remain unchanged.
The only pre-operation controls are credential/cross-user isolation and approval before outward effects. Actual provider or repository command errors are handled when they occur.
MIT