[code-crusher]

Summary

• release: v7.0.0
• feat(ui): grouped tool rows and turn-wide spinner stats
• feat: Claude Code-style request layout and prompt-cache fixes

[matterai-app]

Summary By MatterAI

🔄 What Changed

Refined reasoning stream lifecycle handling in agent core and optimized TUI transcript rendering to exclude replayed reasoning entries, alongside introducing a ctrl+o shortcut to toggle grouped tool row expansion.

🔍 Impact of the Change

Improves terminal chat readability by omitting redundant historical reasoning entries and empowers users to interactively expand or collapse grouped tool calls using ctrl+o.

📁 Total Files Changed

Click to Expand

File	ChangeLog
src/core/agent.ts	Reasoning Finalization Updated comments and state handling for reasoning blocks.
src/ui/App.tsx	Tool Expansion State Added expandToolsRef and ctrl+o shortcut handling to toggle tool groups.
src/ui/components/Header.tsx	Shortcut Display Updated header text with ctrl+o expand shortcut.
src/ui/components/rows.tsx	Row Rendering Cleanup Cleaned up unused row render blocks.

🧪 Test Added/Recommended

Added

• None in this specific diff slice.

Recommended

• Add unit tests verifying ctrl+o correctly toggles tool-group expansion state in the transcript.

🔒 Security Vulnerabilities

• None identified.

[matterai-app]

🧪 PR Review is completed: Release-quality changes overall: the cache-stable message-folding design, tool grouping, and spinner stats are well-reasoned and covered by strong new tests. Two issues found: (1) the new zero-arg buildSystemPrompt() leaves the untouched bench/variants.ts:24 caller (buildSystemPrompt(cwd)) failing typecheck if bench/ is covered by tsconfig; (2) getGitSummary renders '(clean)' when git status fails. Reviewed src/api/aiSdkClient.ts, src/api/llmClient.ts, src/ui/App.tsx, src/ui/components/Spinner.tsx, src/ui/components/rows.tsx, test/agent-context.test.ts, test/tool-group.test.tsx: no issues found.

Skipped files

• CHANGELOG.md: Skipped file pattern
• package-lock.json: Skipped file pattern

⬇️ Low Priority Suggestions (2)

src/prompts/system.ts (1 suggestion)

Location: src/prompts/system.ts (Lines 256-257)

🟠 Build Break Risk

Issue: buildSystemPrompt changed from (cwd: string, options?) to a zero-argument function, but bench/variants.ts:24 still calls buildSystemPrompt(cwd) and is not updated in this PR. If bench/ is covered by tsconfig.json, npm run typecheck (and therefore prepublishOnly) fails with TS2554: Expected 0 arguments, but got 1 — a broken v7.0.0 release build.

Fix: Update bench/variants.ts:24 to let prompt = buildSystemPrompt(), or apply the suggested backward-compatible signature so the legacy caller keeps compiling.

Impact: Prevents a failed typecheck/publish for the release (if bench/ is typechecked).

-  export function buildSystemPrompt(): string {
-    return [roleDefinition, toolGuide, harnessSection].join("\n\n");
+  export function buildSystemPrompt(_cwd?: string): string {
+    return [roleDefinition, toolGuide, harnessSection].join("\n\n");

src/core/agent.ts (1 suggestion)

Location: src/core/agent.ts (Lines 315-315)

🔵 Robustness

Issue: gitOutput returns undefined when a git command fails, so ?? "" maps a failed git status --short to an empty list — which then renders as "(clean)". The old getGitSummary bailed out entirely (empty summary) when git status threw, so a failing status (e.g. a locked index during a concurrent git operation, where git rev-parse still succeeds) now misinforms the model that the working tree is clean.

Fix: Fall back to an explicit "(unavailable)" marker so a failed status is distinguishable from a genuinely clean tree.

Impact: The model no longer receives a false "clean working tree" signal in the rare case git status errors while the branch lookup succeeds.

-  	const status = (gitOutput(cwd, "git status --short") ?? "").split("\n").filter(Boolean)
+  	const status = (gitOutput(cwd, "git status --short") ?? "(unavailable)").split("\n").filter(Boolean)