Agent selection lost outside Orca too: "Agent not found" after a failed reload (2.2.7, MATE Terminal) — same as #62002 #63528

Summary

This is the same defect as #62002 — a session cannot select its configured primary agent and dies with AgentNotFoundError — but it is not Orca-specific: it reproduces with plain opencode running in a normal terminal, MATE Terminal (VTE 0.76.0), no Orca window in that session's chain.

#62002 was closed as not planned on 2026-11-12 06:36 UTC, after the closing note said the configured default_agent "came back as soon as agents/ORACLE.md was restored" and that the behaviour only remained inside Orca's terminal. That explanation does not fit this setup, and the failure is still happening on 2.2.7 outside Orca. Reopening the original was not possible from this account (GraphQL: Could not reopen the issue. (reopenIssue) / REST 422), so this report carries the reproduction instead.

Environment

• opencode 2.2.7 (opencode --version), Linux Mint, x86_64.
• Terminal: MATE Terminal, VTE_VERSION=7600, TERM=xterm-256color, COLORTERM=truecolor, TERM_PROGRAM unset.
• default_agent: "ORACLE" in ~/.config/opencode/opencode.json, with the agent defined inline in the agents block of that same file. There is no ~/.config/opencode/agent/ORACLE.md — so "restore the agent file" is not available as a workaround here.

What happens

The TUI keeps showing the configured agent in the status bar (ORACLE · DeepSeek V4.1 Flash 1017), while the server refuses to select it and drains the turn:

Error: Agent not found: "ORACLE"

Server log, same machine, 2026-11-12 — 9 occurrences today, each within seconds of a failed reload:

level=ERROR message="failed to reload plugins" cause="Cause([Die(TypeError: o is not a function. (In 'o()', 'o' is an instance of Object))])"
level=ERROR message="Failed to drain Session"
  cause="Session.AgentNotFoundError: Agent not found: \"ORACLE\""
    at SessionContext.select
    at SessionRunner.prepareContext
    at SessionRunner.advanceToStep
    at SessionRunner.drain
    at ServerProcess.start

Reload → failure pairs (local time):

A second primary agent defined the same way (ORCHESTRATOR) fails identically at 10:59:59, so it is not specific to one agent name. The same sessions survive until a reload runs; after it, the agent is gone from the registry. The config file was back to a valid, unchanged copy afterwards, and the affected sessions still did not recover the agent — only a restart did.

Expected

• A config file that is momentarily unreadable or invalid — an editor saving it, or a backup restored over ~/.config/opencode/ — must not change the agent a running session uses. The session keeps the selection it already had.
• When the file becomes valid again, the session must recover the configured default_agent by itself. Today the loss is permanent: the file went back to normal and the open windows stayed without their agent until the session was restarted.
• Restoring a backup over the config folder is the concrete trigger: it replaces opencode.json wholesale, every open window loses its configured agent, and a perfectly valid restored file does not bring it back.
• Agents defined inline in opencode.json (not files under agent/) should survive that path — they are never re-read on their own, which is the reproducible core of this report.

Evidence

• Close-up of the error at the top of this report: Error: Agent not found: "ORACLE", marked with 1, with the status bar reading ORACLE · DeepSeek V4.1 Flash 1017 directly below it.
• Full MATE Terminal window with the OpenCode session: attached in the comments.

Related: #62002 (same failure, closed as not planned with an Orca-only explanation).