fix(opencode): show bash commands instead of repeating their output

[brynne8]

Fixes #7307

Problem

With the OpenCode provider, bash tool calls showed the command output as the collapsed row title and the same output again inside the expanded body — the actual command being run was never visible (see screenshot in #7307). Claude and Codex users were not affected.

Root cause

The OpenCode adapter emitted item.completed with detail set to the command's output (part.state.output) while never reading the real invocation at part.state.input.command. For command_execution items, clients fall back to treating detail as the command when no structured command field is present, so:

• collapsed title := command fallback = output
• expanded body := command block + detail block = output twice
• the actual command was dropped entirely

Fix

Adapter-side, one place fixes web/desktop/mobile. The message.part.updated handler now projects command tools into the payload shape clients already render for Claude/Codex:

• detail / data.command carry the command (state.input.command)
• on completion the output moves to data.rawOutput.content, which clients render once in the expanded view
• the unread raw data.state blob is no longer sent (nothing consumed it; it duplicated the whole output over the websocket)

Error states keep surfacing state.error; non-command tools are unchanged.

Testing

• Added an adapter test asserting running/completed projections (OpenCodeAdapter.test.ts)
• Added a client-shape test asserting the OpenCode payload renders as command + single output (session-logic.command-output.test.ts)

Worked on by ox-alpha via opencode.

---

Note

Medium Risk
Changes provider event payloads for command tools, including dropping data.state. Low blast radius but it is a wire-format change for OpenCode tool items.

Overview
Fixes OpenCode bash rows showing stdout as the collapsed title (and again in the expanded body) instead of the actual command (#7307).

OpenCodeAdapter now reads state.input.command for command_execution items. detail/data.command carry the invocation; completed output moves to data.rawOutput.content. Failures still use state.error as detail while keeping data.command. The unused data.state blob is no longer sent.

A client work-log test confirms the new payload renders as command + a single output block.

^{Reviewed by Cursor Bugbot for commit e343705. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Fix OpenCodeAdapter to show bash commands instead of repeating output

• Modifies tool-part event projection in OpenCodeAdapter.ts to set payload.detail to the command string for command_execution items, and to payload.data.rawOutput for the tool output on completion
• Adds commandFromToolInput helper to extract a trimmed command from a tool part's state.input.command
• For error states, payload.detail shows the failure reason while payload.data.command preserves the original command
• Behavioral Change: removes the entire state object from payload.data; consumers expecting data.state must use data.command and data.rawOutput instead

^{Macroscope summarized e343705.}

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 1ba432d1-846c-4f54-aedd-7e7fc861e8bb

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

Want fixes drafted automatically? Bugbot Autofix can create code changes for findings. A team admin can enable Autofix in the Cursor dashboard.

^{Reviewed by Cursor Bugbot for commit f08593b. Configure here.}

[macroscopeapp]

Approvability

Verdict: Skipped

Macroscope did not run approvability analysis for this PR. Macroscope could not determine whether this PR modifies its approvability configuration, so the PR was not approved automatically. A PR that may change the rules that govern approval is never approved automatically.