Skip to content

Troubleshooting

Troubleshooting should start from the symptom the user sees, not from repository internals. A person trying UMMAYA needs to know whether the command is installed, sign-in worked, a public lookup can run, or a protected step correctly stopped.

Maintainer debugging still matters, but it comes after the user path is clear. If the first answer is “run tests” or “inspect git status”, the docs have skipped the reader’s problem.

Use the visible symptom to choose the first check. Do not jump to deep debugging until the simple user path is ruled out.

SymptomFirst checkLikely next step
ummaya command not foundinstall pathrerun installer, Homebrew cask, or npm global install
command starts but cannot sign inFriendliAI login or token statesign in again and confirm provider configuration
first prompt returns no useful resultprompt shape and public adapter availabilitytry a public lookup prompt with a clear place
answer says Mockdomain has shape but no live authorityread Live/Mock/Handoff and treat it as simulation
answer says Handoffnext step needs official authoritycontinue through the official service
session resume failssession ID and local session availabilitycheck the printed resume command and local storage

The table is a triage map, not a proof. If a symptom repeats after the first fix, capture the exact command, visible message, and page or workflow where the failure happened.

If the command is missing, first confirm which installation method you used. The packaged CLI is the user path; source checkout commands are for contributors.

Terminal window
ummaya --version

If the shell cannot find ummaya, reinstall through the chosen package path and open a new shell. If the command exists but fails at startup, record the visible error before trying another installer.

UMMAYA uses FriendliAI/K-EXAONE as the model provider. If sign-in fails, the first question is whether the provider credential exists and the CLI can reach it. A login failure is not an adapter failure and should not be described as a public-service problem.

After fixing login, use a safe public prompt before trying a protected workflow. A good smoke prompt asks for public weather, road, hospital, or safety information with a clear location.

Mock and Handoff are not errors by themselves. Mock means UMMAYA demonstrated a workflow shape without official completion. Handoff means the next step must happen through an official service because UMMAYA lacks a safe callable path.

The recovery is to read the state label and decide what you need next. If you wanted a demo, Mock may be enough. If you wanted a real filing, payment, certificate, identity verification, or record change, Handoff is the honest result unless live authority is configured.

Maintainers can inspect generated docs, tests, IPC frames, adapter schemas, and TUI captures after the user symptom is preserved. The debugging note should keep the original symptom visible: command, prompt, expected state, actual state, and whether the failure happened in install, provider, retrieval, permission, adapter execution, or rendering.

Do not replace a user-facing failure with internal shorthand. “Adapter error” is not enough. Say which adapter, which mode, which primitive, and which stop reason were involved.

If none of the user checks works, collect the minimum useful report: operating system, install method, ummaya --version, prompt used, visible state label, and exact stop message. That report gives maintainers enough context without asking the user to understand the whole repository.