Troubleshooting
For most failures, the agent should investigate and fix the problem. You do not need to interpret test output, choose another command, or repair HelmIt files yourself. Use this page when you are unsure how to resume the conversation or when an action from you is actually needed.
HelmIt does not appear in the client
Section titled “HelmIt does not appear in the client”Ask the agent to check whether helmit@helmit is installed and available in
the Claude Code or Codex client you are using:
“Check why the HelmIt skills do not appear in this client and help me load the plugin.”
After an update, Claude Code may need to reload plugins or start a new session. In Codex Desktop, start a new task: an older task may keep using the version it already loaded. See Installation and getting started to install or update the plugin.
The session ended before the work was complete
Section titled “The session ended before the work was complete”Open the same project and invoke /helmit:next. It checks current records and
shows how to continue. If there are uncommitted changes, the agent should
inspect them before deciding what remains; there is no need to start over.
“Resume this project from where it stopped. Check the completed work and pending changes before continuing.”
See Handoffs for what HelmIt uses when resuming. In
Codex CLI, select the helmit:next skill with $ or /skills.
The agent stopped after a failure
Section titled “The agent stopped after a failure”A failed test, commit, task check, or phase closure should be diagnosed by the agent. Ask it to show the cause, fix what is in scope, and rerun the affected proof:
“Investigate the failure, fix its cause without discarding work already done, and rerun the necessary check.”
If the agent only repeats the error or asks you to resolve a technical detail, ask it to explain which decision or access depends on you. Failed proof should not be ignored to close a task. See Tests for what each check demonstrates.
The agent needs a decision from you
Section titled “The agent needs a decision from you”Some stops are not technical: a product or architecture choice, a criterion
marked [human], external access, or an action you have not yet authorized.
Ask for one clear question with options and their consequences. Once you
answer, the agent can continue from where it stopped.
“Explain what depends on my decision, the options, and what changes with each one. Preserve the work while we decide.”