Activity

1. opencode-agent commented on Oct 18, 2026

   

   opencode-agentBot

   on Oct 18, 2026 – with OpenCode Agent

   Contributor

   More actions

   Thanks for the clear report and for linking the earlier issues. You're right that the v2 root command has no --agent flag: in packages/cli/src/commands/commands.ts only run and mini define agent and model, but the TUI already reads args.agent, so the flag just never reaches it.

   I think this is a duplicate of #56218, which is still open and asks for both -m and --agent to work again on the root command, as they did in v1. Possible duplicates and related items, ordered by likelihood:

   • v2: opencode -m provider/model no longer accepted on root command (worked in v1) #56218: open request to restore -m/--model and --agent on the root TUI command
   • cli: restore --agent (and --model) on the full-screen TUI in v2 #53728: your earlier report of the same missing --agent flag (reproduced, then closed automatically because of the template)
   • fix(cli): pass --model through to the TUI entry #48163: open PR that adds --model to the root command; --agent would need the same small change, and you've already asked there about combining them

   The bot will close this issue as a duplicate in 1 day unless you reply explaining how your problem is different, for example if you'd like --agent tracked separately from #56218.
2. opencode-agent

   added

   pending close

   triaging

   on Oct 18, 2026
3. ShreeMulay commented on Oct 18, 2026

   

   ShreeMulay

   on Oct 18, 2026

   Author

   More actions

   #56218 is primarily tracking --model, and its active implementation PR (#48163) is model-only. I would like to keep this issue as the linked bug for the independent --agent patch. I will keep that PR scoped to --agent and coordinate with #48163 so the two changes do not conflict.
4. opencode-agent commented on Oct 18, 2026

   

   opencode-agentBot

   on Oct 18, 2026 – with OpenCode Agent

   Contributor

   More actions

   Thanks, that makes sense. #56218 and its PR #48163 only add --model, so I'll keep this issue open separately, as the request for opencode --agent <name> to start the full-screen TUI with that primary agent, like mini and run already do. I've withdrawn the duplicate suggestion, so the bot won't close this issue.

   Maintainers will consider it. Related:

   • v2: opencode -m provider/model no longer accepted on root command (worked in v1) #56218: same request for -m/--model on the root command
   • fix(cli): pass --model through to the TUI entry #48163: open PR that adds --model to the root command; an --agent change would touch the same root command options
   • cli: restore --agent (and --model) on the full-screen TUI in v2 #53728: earlier report of the missing --agent flag, reproduced on v2.0.20
5. opencode-agent

   removed

   triaging

   pending close

   on Oct 18, 2026
6. chrisLytsoft commented on Oct 21, 2026

   

   chrisLytsoft

   on Oct 21, 2026

   More actions

   Confirming from our side: #57778 is a duplicate of this issue with the identical root cause (v2's top-level parser has no --agent; the v1 launch form dies in a bare ShowHelp ~3 ms after cli starting; the same flags still work on mini/run). Please treat #56842 as the canonical tracking issue for this bug from our side — we will not reply on #57778, so it can close as a duplicate of this issue.

   Adding the headless detail the triage on #57778 asked for, which I don't think is stated above yet:

   Headless silent-failure amplifier. Scheduled wrappers that launch opencode detached (Start-Process from a PowerShell Task Scheduler action) exit 0 regardless of the CLI failure, so the scheduler reports success (LastTaskResult 0) while nothing actually runs. Our case: after the npm v1→v2 upgrade on 2026-10-16, daily scheduled sessions failed silently from 2026-10-16 to 2026-10-21 — several days of automation "succeeding" while zero sessions were created. Detection required correlating session history against ~/.local/share/opencode/log/opencode.log (the run=… ShowHelp pair quoted on #57778); the exit codes and scheduler state were green throughout.

   Diagnostic gap suggestion. The bare ShowHelp failure names no flag — nothing distinguishes "you passed an unknown flag" from any other help output, which is exactly the case for headless callers. An "unknown flag: --agent"-style error message (or a non-zero, distinguishable exit code for argument-parse failures) would surface this entire class immediately instead of after days of silent failure.

   Workaround confirmed for headless use: opencode mini --agent <name> --prompt "<text>" launches and consumes the prompt correctly in the scheduled, detached context (verified with process, log, and session-record evidence).