[trs-80]

Summary

This PR removes an unused per-agent hook-status IPC surface and consolidates two duplicated agent-registry behaviors without changing the exported desktop configuration shape:

1. Removes the unused window.api.agentHooks.*Status IPC/preload/web surface. Existing agent launch/status tracking remains intact.
2. Authors launchCmd and expectedProcess as optional source fields and resolves both from detectCmd once at module initialization.
3. Derives the mobile agent order, labels, and picker helpers from the shared TUI registry while keeping mobile-only favicon domains local.
4. Updates the web preload composition contract test to match the intentionally removed namespace.

The branch was rebased onto current main during review. Final diff: 12 files, +93 / -592.

Why

These facts were duplicated across desktop and mobile and could drift when agents were added or changed. The final implementation keeps one shared source for stable agent metadata while preserving existing consumer contracts and leaving broader registry consolidation out of scope.

Compatibility and risk

• No RPC, remote-wire, persisted-state, route, pairing, or protocol changes.
• No new dependencies, subprocesses, timers, listeners, polling, or network behavior.
• Mobile IDs, ordering, labels, disabled-agent semantics, and provider-neutral behavior are preserved.
• No platform-specific path, shell, shortcut, native-module, SSH, relay, or folder-workspace behavior changed.
• Registry resolution is a bounded 36-entry module-initialization pass; mobile membership checks use Object.hasOwn.

Verification

• pnpm typecheck
• Mobile TypeScript check
• pnpm lint
• pnpm build
• Focused root and mobile unit/integration suites covering the changed registries, preload composition, and hook IPC
• iOS Metro export: 4,638 modules bundled
• Same-model senior review loop: clean after fixing the stale web preload composition expectation
• Opening and closing release-readiness reviews: no proven P0/P1/P2 findings

Electron rendered QA

Validated this exact worktree with isolated user data and Playwright CDP. The real Create Worktree and New Tab agent pickers rendered the expected detected-agent labels/order; Codex was selected and launched successfully. window.api.agentHooks was absent, while agentStatus and launch/foreground tracking remained present. Renderer console had no errors.

Native mobile QA

Native iOS validation was attempted on iPhone 17 Pro through the Orca-managed emulator and this worktree's Metro bundle. It was blocked before the agent picker by an existing home-screen hosts.toSorted render error outside this PR's diff. No web render was substituted for native evidence; the changed catalog remains covered by focused tests and the successful iOS Metro export.

AI disclosure

The original implementation was AI-assisted. The final rebased head was reviewed by multiple independent agents plus a same-model fix/review loop, with only one meaningful regression found and fixed: the stale web preload surface assertion.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 682b5a7d-cdeb-48b0-b38d-710c5c9ea669

📥 Commits

Reviewing files that changed from the base of the PR and between bc98655 and feeea7d.

📒 Files selected for processing (12)

• mobile/src/tasks/mobile-agent-catalog.test.ts
• mobile/src/tasks/mobile-tui-agents.ts
• src/main/ipc/agent-hooks.test.ts
• src/main/ipc/agent-hooks.ts
• src/preload/api-types.ts
• src/preload/api/agent-hook-api.ts
• src/preload/index.ts
• src/renderer/src/web/preload-api/web-agent-hooks-api.ts
• src/renderer/src/web/web-preload-api-composition.test.ts
• src/renderer/src/web/web-preload-api.ts
• src/shared/tui-agent-config.test.ts
• src/shared/tui-agent-config.ts

💤 Files with no reviewable changes (7)

• src/renderer/src/web/web-preload-api-composition.test.ts
• src/preload/api/agent-hook-api.ts
• src/preload/index.ts
• src/renderer/src/web/preload-api/web-agent-hooks-api.ts
• src/main/ipc/agent-hooks.test.ts
• src/renderer/src/web/web-preload-api.ts
• src/main/ipc/agent-hooks.ts

🚧 Files skipped from review as they are similar to previous changes (5)

• src/preload/api-types.ts
• src/shared/tui-agent-config.test.ts
• mobile/src/tasks/mobile-tui-agents.ts
• mobile/src/tasks/mobile-agent-catalog.test.ts
• src/shared/tui-agent-config.ts

Included review availability: Your plan provides up to 10 included reviews per hour; 9 remain after this review.

---

📝 Walkthrough

Walkthrough

The change centralizes TUI agent configuration defaults and reuses shared ordering, labels, validation, enablement, and selection logic in mobile code. Mobile catalog tests now validate shared configuration and display names. Provider-specific hook status IPC handlers, preload types, preload bindings, and web fallback APIs were removed. Generic agent status handling remains. Configuration tests cover defaults and explicit command overrides.

🚥 Pre-merge checks | ✅ 3 | ❌ 2

❌ Failed checks (2 warnings)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 16.67% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 6 functions across 5 files.	Write docstrings for the functions missing them to satisfy the coverage threshold.
Description check	⚠️ Warning	The description is detailed and on-topic. It explains the changes, rationale, compatibility impact, verification, visual evidence, mobile limitations, and AI use. However, it omits the required linked…	Add a valid issue reference under Linked Issue, such as Fixes #123``. Include or explicitly address the required ELI5, What Changed, Visual Proof, Testing, Review, Agent skill upstream boundary, Notes, and Checklist sections. Confirm the re…

✅ Passed checks (3 passed)

Check name	Status	Explanation
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Title check	✅ Passed	The title clearly summarizes the two primary changes: removing unused agent hook IPC and deriving shared agent defaults.

Full details: Description check

Explanation

The description is detailed and on-topic. It explains the changes, rationale, compatibility impact, verification, visual evidence, mobile limitations, and AI use. However, it omits the required linked issue, does not use several required template sections, and does not include the checklist confirmations.

Resolution

Add a valid issue reference under Linked Issue, such as Fixes #123``. Include or explicitly address the required ELI5, What Changed, Visual Proof, Testing, Review, Agent skill upstream boundary, Notes, and Checklist sections. Confirm the required checks and manual testing status.

---

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.}

[brennanb2025]

Thanks for this — the audit behind it (recent agent additions touching 27–47 files, a third of those edits re-declaring facts already in TUI_AGENT_CONFIG) matches what we've been running into, and three of the four cleanups here are straightforwardly good.

The ask: drop b3d5bf8e47 and keep the other three.

Why just that one

Main moved under this PR while it was open. src/shared/agent-title-evidence.ts landed in #16148, and it doesn't deduplicate the title chain — it replaces the way title identity is decided. collectAgentTitleEvidence sorts signals into three classes (vendor marker / anchored name / free-text name) and ranks by class, so a shuffle of its registry cannot change the answer. #16157 then put the whole title signal behind hook, process, and launch evidence.

AGENT_TITLE_MATCHERS is the opposite contract. It makes "first matcher wins" explicit and public, and preserves the order faithfully — which is exactly right as a refactor, and exactly the thing being removed. Concretely:

• It keeps all 8 defects pinned by the characterization suite in test(agent-status): characterize title-derived agent identity before the resolver change #16144. Four are real recorded panes: Switch Claude and Codex off load balance… - grok is a Grok pane that resolves to Codex, because Codex is checked first and the task text mentions it.
• The table's own test asserts list position (at(-1) must be claude), so ordering becomes a documented invariant future changes have to protect.
• It leaves two live chains in src/shared with opposite semantics — one order-dependent, one deliberately not.

The duplicate chains do still need to go. They come out as consumers migrate onto the evidence parser (#16237 is the first), so that cleanup arrives without publishing an ordered table in the meantime.

The three to keep

• 39ea6ad7a0 — dead agentHooks:*Status IPC. Verified: zero renderer/CLI/mobile callers on current main, and the surface had already drifted (kimiStatus existed in main and preload but not in the type or the web stub).
• d1f7975507 — launchCmd/expectedProcess defaulting to detectCmd. Resolved table verified toEqual the previous one.
• 18e26e053f — mobile registry derived from src/shared. Verified labels identical to TUI_AGENT_DISPLAY_NAMES across all 36 agents, the shared helpers all exist, and the import is Metro-safe.

Those are real reductions we'd otherwise have to write ourselves, and none of them are affected by the identity work.

Practical note

The branch needs a rebase regardless — git merge-tree now reports conflicts in terminal-title-agent-type.ts and agent-hooks.ts against current main. If it's easier, drop b3d5bf8e47 during that rebase and the remaining three should apply cleanly.

Happy to take the rebase off your hands if you'd prefer — say the word and we'll land the three with credit to you.

[coderabbitai]

Note

GitHub couldn't provide a complete incremental comparison for this pull request, so CodeRabbit is performing a full review instead. This review may take a little longer.

[trs-80]

Rebased onto current main (cc4801320a) with b3d5bf8e47 dropped — agreed that publishing an ordered matcher table has no place once agent-title-evidence owns title identity, and the duplicate chain can come out through the #16237 migration path instead.

The three keepers are now:

• 604c2c151f — dead agentHooks:*Status IPC. The only conflict of the rebase was here: main's feat(agent-status): add an order-independent title evidence parser #16148-era extraction of the agentStatus:drop* listeners into registerAgentStatusRowTeardownIpcHandlers() crossed with this commit's import removals. Resolved to current main's structure minus the 14 per-agent hook-service imports; no agentHooks: handler survives and the hook services have no other reference in the file.
• 0be1fc4458 — launchCmd/expectedProcess defaulting to detectCmd (applied clean).
• 4930851ab6 — mobile registry derived from src/shared (applied clean).

Re-verified on the new base: full pnpm typecheck (node/cli/web/e2e) clean, agent-hooks + tui-agent-config suites pass, and the mobile package typechecks with all 3,681 tests passing.

🤖 Generated with Claude Code

[coderabbitai]

Note

GitHub couldn't provide a complete incremental comparison for this pull request, so CodeRabbit is performing a full review instead. This review may take a little longer.

[brennanb2025]

Review status: ready after required CI

Reviewed final head feeea7d6b09f69f1ea6781c843ee40cbf05fbb3f and pushed the rebased branch to the contributor PR.

Changes made during review

• Rebased onto current main and resolved the web preload split within the PR's existing scope.
• Fixed one meaningful regression: the exact web preload composition test still expected the intentionally removed agentHooks namespace (feeea7d6b0).
• Updated the PR title/body to match the actual 12-file diff and added rendered Electron evidence.

Review result

• Same-model senior review/fix loop: CLEAN after the test correction.
• Opening and closing readiness checklists: PASS, with zero proven P0/P1/P2 findings; counsel was correctly skipped because there were no candidates.
• No scope expansion, dependency/supply-chain change, hot-path regression, new polling/listeners/timers, wire/persistence change, or Windows/Linux/SSH/folder-workspace/provider-specific regression found.

Verification

• Root and mobile typechecks passed.
• Lint and build passed.
• 117 focused root tests and 17 focused mobile tests passed.
• iOS Metro export passed (4,638 modules).
• git diff --check passed.
• Electron Playwright CDP validation passed on this exact worktree: the real Create Worktree and New Tab agent pickers rendered correctly, Codex selected/launched successfully, window.api.agentHooks was absent, and remaining agentStatus plus launch/foreground tracking stayed intact. Screenshots are attached in the PR description.

Remaining gap

Native iOS picker validation was attempted through the Orca-managed emulator, but the app hit an existing hosts.toSorted home-screen render error in mobile/src/home/use-mobile-home-data.ts:131 before reaching the picker. That file is outside this PR's diff, so this is not a PR finding; native QA is reported as blocked rather than replaced with web evidence.

GitHub now reports the rebased PR as mergeable. Required checks have not populated for this pushed SHA yet, so the landing recommendation is: land after required CI passes.