[Tauri-EPO]

ELI5

orca terminal create --command "codex" starts Codex in a new pane but never tells the pane "this is Codex". Orca only figures that out later, when Codex changes the terminal title. If a script runs worker-start right after create, the paste arrives before that happens, gets the generic "wait a fixed delay, then press Enter" treatment, and stalls. Now create recognizes the agent from the command line and labels the pane immediately.

What Changed

• src/cli/handlers/terminal.ts: terminal create runs the command through recognizeAgentProcessFromCommandLine and passes launchAgent to terminal.create when an interactive agent is recognized. Headless one-shots stay unlabeled: the shared recognizer already drops claude -p, and the existing Codex classifier (shouldUseRendererBackedCodexTerminal) rules out codex exec and the other non-interactive subcommands.
• src/cli/handlers/terminal.test.ts: codex, codex --yolo, claude --model … set launchAgent; codex exec …, claude -p …, npm run dev and no command leave it unset.

Why

#17291 traced it against source: pty.launchAgent starts null, pty.foregroundAgent is populated only after an OSC title observation, and createAgentPromptRenderGate returns the real echo gate only for a recognized agent. A worker-start --terminal <handle> chained immediately after terminal create --command "claude …" therefore takes the open-loop delay and reports agent_prompt_stalled; any pause of a couple of seconds between the two commands makes it pass. The terminal.create RPC already accepts launchAgent (the e2e fixtures use it); the CLI just never sent it. This is the fix the issue proposed, gated on the Codex classifier instead of the renderer-backed flag so remote clients get it too.

Linked Issue

Closes #17291

Visual Proof

N/A — CLI-only; the pane created by terminal create --command codex now carries launchAgent: codex from the start.

Testing

• I manually tested these changes locally
• Automated tests added/updated, or explained why not below

Ran on Windows 11:

pnpm test src/cli/handlers/terminal.test.ts   # 15 passed
pnpm tc:cli
oxlint + oxfmt on the changed files

AI Disclosure

Written with Claude Code (Claude Fable 5.1) under direction of @Tauri-EPO.

Review

Agent skill upstream boundary

• Not applicable, or this change follows docs/reference/agent-skill-sharing-upstream-boundary.md and copies or mechanically translates no upstream skill-installer source, tests, fixtures, registry entries, path tables, comments, or documentation.

Notes

Sends an already-supported optional RPC field, so old hosts ignore it. No platform-specific behavior; the recognizer handles .exe/.cmd basenames.

Checklist

• This PR is small and focused
• I explained what changed and why (including ELI5)
• Before/after screenshots or videos attached for UI changes, or N/A with reason
• Self-reviewed for correctness, security, and performance
• Cross-platform, SSH/remote, and path/shortcut impact considered (or N/A)
• pnpm lint, pnpm typecheck, pnpm test, and pnpm build pass (or CI will cover; local preferred) — tc:cli, the handler tests and oxlint on changed files pass locally; full lint/build left to CI

🤖 Generated with Claude Code

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Team

Run ID: c2c5f33d-e101-4c68-87dd-eec0af33d6ca

📥 Commits

Reviewing files that changed from the base of the PR and between 7c56d16 and c42dc01.

📒 Files selected for processing (2)

• src/cli/handlers/terminal.test.ts
• src/cli/handlers/terminal.ts

🚧 Files skipped from review as they are similar to previous changes (2)

• src/cli/handlers/terminal.test.ts
• src/cli/handlers/terminal.ts

Included review availability: Your plan provides up to 10 included reviews per hour; 7 remain after this review.

---

📝 Walkthrough

Walkthrough

The terminal create handler now includes launchAgent only for interactive claude and codex commands. Other commands leave the field undefined. Tests cover version, additional agent, and one-shot commands.

Merge Risk: ⚪ Minimal · up to c42dc

This localized CLI change labels recognized interactive agents when terminals are created, improving immediate command handling without introducing an actionable merge-blocking risk; it is merge-ready after normal checks and review.

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 1 functions across 2 files.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and concisely describes the main change: inferring launchAgent for interactive agent commands during terminal creation.
Description check	✅ Passed	The description covers the required sections, explains the problem and solution, links issue #17291, documents CLI-only behavior, and records testing. The unchecked manual and full-suite checks are ac…
Linked Issues check	✅ Passed	The implementation meets issue #17291 by recognizing interactive Claude and Codex commands during terminal creation and passing launchAgent immediately. Tests also verify that headless and unrelated c…
Out of Scope Changes check	✅ Passed	The changes are limited to terminal creation launchAgent inference and related test coverage. They directly support the linked issue and PR objectives.

Full details: Description check

Explanation

The description covers the required sections, explains the problem and solution, links issue #17291, documents CLI-only behavior, and records testing. The unchecked manual and full-suite checks are acceptable because automated tests and CI coverage are documented.

Full details: Linked Issues check

Explanation

The implementation meets issue #17291 by recognizing interactive Claude and Codex commands during terminal creation and passing launchAgent immediately. Tests also verify that headless and unrelated commands remain unlabeled.

• Fix all pre-merge checks with AI

---

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.}

[coderabbitai]

Actionable comments posted: 1

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Team

Run ID: c7175775-a997-40a5-bac9-e54aa30db327

📥 Commits

Reviewing files that changed from the base of the PR and between 7314ada and 7c56d16.

📒 Files selected for processing (2)

• src/cli/handlers/terminal.test.ts
• src/cli/handlers/terminal.ts

Included review availability: Your plan provides up to 10 included reviews per hour; 8 remain after this review.

[pullfrog]

ℹ️ No critical issues — the fix is correct for its stated goal and the tests actually fail without it. One rough edge worth a look before merge.

Reviewed changes

• terminal create infers launchAgent — src/cli/handlers/terminal.ts runs the --command through recognizeAgentProcessFromCommandLine and forwards launchAgent to terminal.create, with an interactivity gate only for codex (via shouldUseRendererBackedCodexTerminal).
• Handler tests — new describe('terminal create CLI') asserts launchAgent is set for codex, codex --yolo, claude --model … and unset for codex exec …, claude -p …, npm run dev, and no command. Exact toMatchObject / not.toHaveProperty assertions, so they genuinely fail without the behavior.

ℹ️ Nitpicks

• The codex interactivity gate also drops a genuinely interactive codex launch when it arrives via a node entrypoint (node node_modules/@openai/codex/…): the shared recognizer resolves that to codex via recognizeNodeScriptEntrypoint, but shouldUseRendererBackedCodexTerminal returns false for a node first token, so launchAgent is left unset. Edge-case only (CLI users type codex, not the packaged entrypoint), so this is informational rather than blocking.

  ^{｜ Fix all ➔ ｜ Fix 👍s ➔ ｜ View workflow run ｜ Using DeepSeek Pro (free via Pullfrog for OSS) ｜ 𝕏}

[pullfrog]

✅ No new issues found. The previous asymmetric-gating concern is resolved: launchAgent is now restricted to claude/codex and requires shouldUseRendererBackedInteractiveTerminal(command), so non-interactive agent commands (gemini, opencode run, aider --message) and claude --version stay unlabeled. The added test cases genuinely fail without this gate.

Reviewed changes

• Tightened the interactivity gate — src/cli/handlers/terminal.ts now requires both (recognizedAgent === 'claude' || recognizedAgent === 'codex') and shouldUseRendererBackedInteractiveTerminal(command) before inferring launchAgent, replacing the codex-only gate.
• Expanded negative tests — src/cli/handlers/terminal.test.ts adds claude --version, gemini, opencode run "summarize", and aider --message "fix" to the "leaves launchAgent unset" cases.

  ^{｜ View workflow run ｜ Using DeepSeek Pro (free via Pullfrog for OSS) ｜ 𝕏}