Harden chat protocol parsing and submit flow

[TraderSamwise]

Summary

• harden Claude/Codex terminal transcript parsing for prompts, responses, status rows, and startup chrome
• route tmux-backed GUI sends through the carriage-return prompt submit path
• add composer protocol tests and attachment URL rendering coverage
• keep context capture focused on parsed assistant responses instead of terminal chrome

Verification

• YARN_IGNORE_ENGINES=1 yarn vitest src/context/context-bridge.test.ts src/agent-output-parser.test.ts src/multiplexer/session-runtime-core.test.ts --run
• YARN_IGNORE_ENGINES=1 yarn typecheck
• YARN_IGNORE_ENGINES=1 yarn lint
• YARN_IGNORE_ENGINES=1 yarn build
• YARN_IGNORE_ENGINES=1 git push pre-push hook: typecheck, lint, 648 tests
• Chrome MCP live smoke: Codex CODEX_PROTOCOL_OK and Claude CLAUDE_PROTOCOL_OK_2 round trips

Summary by CodeRabbit

• Bug Fixes

  • Fixed image URL normalization to properly handle relative paths and absent URLs
  • Improved message sending eligibility logic for consistent send enablement

• Improvements

  • Refined agent response parsing to better classify and extract content, status, and prompt sections
  • Enhanced multi-line message submission handling for terminal-backed sessions
  • Improved message composition input validation and send-text computation

• Tests

  • Added and extended tests to cover composer behavior, parsing, and session input flows

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
app	Ready	Preview, Comment	Jun 2, 2026 8:07am

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: bd8008fe-f3bf-4395-b7f7-a643604bba80

📥 Commits

Reviewing files that changed from the base of the PR and between 78734c3 and bc9250d.

📒 Files selected for processing (2)

• src/agent-output-parser.test.ts
• src/agent-output-parser.ts

🚧 Files skipped from review as they are similar to previous changes (1)

• src/agent-output-parser.ts

---

📝 Walkthrough

Walkthrough

Adds a composer protocol and integrates it into AgentChatScreen and tmux send flow; enhances parseAgentOutput to better classify Codex/Claude UI and preserve trailing prompts; context bridge now uses parser for live snapshots; normalizes image attachment URLs and adds tests.

Changes

Message Composition & Sending

Layer / File(s)	Summary
Composer protocol contracts and helpers app/lib/composer-protocol.ts, app/lib/composer-protocol.test.ts	Exports composer types and helpers: normalizeComposerDraft, shouldSubmitComposerKey, and getComposerSendText. Tests validate draft trimming, Enter-key rules, and conditional send-text logic.
Chat screen message eligibility and submission app/components/screens/AgentChatScreen.tsx	AgentChatScreen imports composer helpers, computes composerSendText for gating and submission, delegates web key handling to shouldSubmitComposerKey, and ties Send button disabled state to composerSendText.
Tmux prompt submission normalization src/multiplexer/session-runtime-core.ts, src/multiplexer/session-runtime-core.test.ts	sendAgentInput now normalizes the prompt, writes the normalized draft to the tmux transport, and awaits waitForTmuxPromptSubmit. Test covers carriage-return prompt submission with fake timers.

Agent Output Parsing & Context Extraction

Layer / File(s)	Summary
Enhanced agent output parsing logic src/agent-output-parser.ts, src/agent-output-parser.test.ts	Adds Codex UI detection and expands isStatusLine heuristics (MCP startup, progress bullets), treats Codex UI as meta before prompts or status after, tightens prompt/empty-line handling, and removes trailing prompt stripping. Tests cover Codex/Claude status parsing, wrapped prompts, and trailing prompt preservation.
Context bridge live snapshot extraction src/context/context-bridge.ts, src/context/context-bridge.test.ts	ContextBridge now parses pane text with parseAgentOutput and selects the last response block for snapshotTurn. Tests updated to ensure chrome/progress rows are excluded and only last real response is recorded.

Message Display Improvements

Layer / File(s)	Summary
Image URL path normalization and null handling app/components/MessageBlock.tsx, app/components/MessageBlock.test.ts	resolveImageUrl prepends a leading / for non-absolute contentUrl values before joining with getServiceUrl(endpoint) and returns null when contentUrl is missing. Tests assert normalized local URLs and null behavior.

Sequence Diagram(s)

sequenceDiagram
  participant User
  participant AgentChatScreen
  participant ComposerProtocol
  participant SessionRuntime
  participant TmuxTransport
  User->>AgentChatScreen: type text / press Enter
  AgentChatScreen->>ComposerProtocol: normalizeComposerDraft(draft)
  ComposerProtocol-->>AgentChatScreen: normalizedDraft or null
  AgentChatScreen->>ComposerProtocol: getComposerSendText(state)
  ComposerProtocol-->>AgentChatScreen: composerSendText or null
  AgentChatScreen->>ComposerProtocol: shouldSubmitComposerKey(event)
  ComposerProtocol-->>AgentChatScreen: true/false
  AgentChatScreen->>SessionRuntime: handleSendMessage(composerSendText)
  SessionRuntime->>TmuxTransport: sendAgentInput(normalized prompt)
  TmuxTransport-->>SessionRuntime: waitForTmuxPromptSubmit result
  SessionRuntime-->>AgentChatScreen: acceptance result

Loading

sequenceDiagram
  participant TmuxPane
  participant ContextBridge
  participant OutputParser
  participant ContextWatcher
  TmuxPane->>ContextBridge: capturePaneSnapshot(normalized text)
  ContextBridge->>OutputParser: parseAgentOutput(pane text)
  OutputParser-->>ContextBridge: parsed blocks (meta/prompt/status/response)
  ContextBridge->>OutputParser: select last response block
  OutputParser-->>ContextBridge: final response text
  ContextBridge->>ContextWatcher: snapshotTurn = response text
  ContextWatcher-->>ContextBridge: snapshot recorded

Loading

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~25 minutes

Possibly related PRs

• TraderSamwise/aimux#52: Related changes touching AgentChatScreen send eligibility and tmux send behavior.
• TraderSamwise/aimux#8: Related earlier changes to tmux prompt submission and awaiting prompt acceptance.

Poem

🐰 I trimmed the draft and watched the keys,

Sent prompts that waited for tmux replies,
Parsers learned to skip the chrome and spinners,
Images found their proper path again,
A rabbit cheers—careful, neat, and spry.

🚥 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%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The PR title accurately summarizes the main changes: hardening chat protocol parsing and the message submit flow, which are the core objectives reflected throughout the changeset.
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.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch codex/chat-protocol-robustness

---

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

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@src/agent-output-parser.ts`:
- Around line 82-86: The current bullet-line regexes treat any bullet containing
"for <n>s", an ellipsis, or a trailing parenthetical as a status, which swallows
normal assistant replies; narrow these patterns in agent-output-parser by
changing the checks that use trimmed (the /^•\s?\S.*\bfor \d+s\b/ and
/^[*✻✽✶]\s+\S.*(?:\bfor \d+s\b|\.\.\.|…|\(.+\))$/ tests) to only match
well-known short status phrases — e.g., require the bullet to start with a
capitalized status keyword
(Starting|Started|Stopped|Running|Restarting|Added|Removed|Updated) or to be a
very short token (limit to 1–3 words) when paired with a
timing/ellipsis/parenthetical, and restrict parenthetical matches to explicit
timing formats like \(\d+(?:ms|s)\); update those regexes accordingly so
ordinary sentences like "I'll wait for 5s..." or "Added tests (2 files)" no
longer qualify as status lines.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 99f57741-8317-4842-8224-2b062bf3c106

📥 Commits

Reviewing files that changed from the base of the PR and between ae9a215 and 78734c3.

📒 Files selected for processing (11)

• app/components/MessageBlock.test.ts
• app/components/MessageBlock.tsx
• app/components/screens/AgentChatScreen.tsx
• app/lib/composer-protocol.test.ts
• app/lib/composer-protocol.ts
• src/agent-output-parser.test.ts
• src/agent-output-parser.ts
• src/context/context-bridge.test.ts
• src/context/context-bridge.ts
• src/multiplexer/session-runtime-core.test.ts
• src/multiplexer/session-runtime-core.ts