Enhance tool evidence display and improve light mode contrast

[Szowesgad]

This pull request introduces several improvements to tool execution handling, image attachment processing, and hotkey gating logic in the agent/controller pipeline. The changes focus on making tool names more user-friendly, improving the handling and error reporting of image attachments, and refining the logic for hotkey events during agent turns. Comprehensive unit tests are added for all new logic.

Tool Execution and Labeling Improvements:

• Added friendly_tool_name and tool_running_status functions to map raw tool identifiers to concise, human-readable labels and provide user-friendly status messages during tool execution. Updated tool execution/result handling to use these labels and avoid showing raw wire names in the conversation timeline. [1] [2]
• Updated logging to include both raw and friendly tool names for debugging, and improved the grouping of tool activity in the UI.
• Added comprehensive unit tests to ensure tool name mapping and status messages are correct and regression-proof.

Image Attachment Handling:

• Implemented build_image_attachments_from_text to parse, load, and cap image attachments referenced in outgoing messages, ensuring images are properly forwarded as vision input and missing/unreadable images are reported to the user. [1] [2]
• Added tests for image attachment parsing, loading, overflow capping, and error reporting.

Agent/Hotkey Gating Logic:

• Refined hotkey gating logic to allow assistive "Talk Anytime" starts while an agent turn is in progress, but continue to block non-assistive dictation starts. Updated should_block_hotkey_during_agent_send and related controller logic for clarity and testability. [1] [2] [3]

Other Improvements:

• Skipped empty image content blocks when building input items for the OpenAI provider, preventing request rejection due to empty image data.
• Minor logging and import updates to support new features. [1] [2]

These changes collectively improve user experience, reliability, and maintainability of the agent/controller interaction pipeline.## Summary

What changed, and why?

User Impact

What does this improve or prevent for a real CodeScribe user?

Runtime Impact

• Recording, hotkeys, overlays, settings, or release packaging changed
• Public docs, onboarding, install path, or metadata changed
• No runtime behavior changed

Verification

• cargo fmt --all
• cargo clippy -- -D warnings
• cargo test
• make semgrep

Add targeted commands, screenshots, recordings, or release-artifact checks here:

Release Notes

Should this appear in CHANGELOG.md?

• Yes, added
• No, internal-only

[gemini-code-assist]

Code Review

This pull request introduces several major enhancements to CodeScribe, including an onboarding lane chooser (Basic vs. Agentic), an Agentic-lane readiness probe, grouped tool activity in the voice chat timeline, Light Mode semantic contrast fixes, and a post-LLM lexicon pass to prevent protected term corruption. The review comments correctly identify opportunities to improve Rust stable compatibility by avoiding unstable let-chains, optimize string truncation performance to prevent O(N) traversal on large inputs, and fix a bug where missing image files are incorrectly reported as valid vision inputs.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[gemini-code-assist]

Using let-chains (&& with let) is currently an unstable nightly feature in Rust. To ensure compatibility with stable Rust compilers, it is safer to use nested if let statements.

Suggested change

	if let Some(idx) = state.active_tool_activity_index
	&& state
	.messages
	.get(idx)
	.map(|msg| msg.role == ChatRole::ToolActivity)
	.unwrap_or(false)
	{
	return idx;
	}
	if let Some(idx) = state.active_tool_activity_index {
	if state
	.messages
	.get(idx)
	.map(|msg| msg.role == ChatRole::ToolActivity)
	.unwrap_or(false)
	{
	return idx;
	}
	}

[gemini-code-assist]

Calling text.chars().count() traverses the entire string to count characters, which is highly inefficient ($O(N)$) for pathological or extremely large inputs (e.g., pasted base64 blobs or giant log files). Since we only care if the string exceeds MAX_BUBBLE_DISPLAY_CHARS, we can use char_indices().nth(MAX_BUBBLE_DISPLAY_CHARS) to find the truncation boundary in $O(\text{MAX})$ time, completely avoiding traversing the rest of the string.

Suggested change

	let capped = if text.chars().count() > MAX_BUBBLE_DISPLAY_CHARS {
	let mut t: String = text.chars().take(MAX_BUBBLE_DISPLAY_CHARS).collect();
	t.push_str("\n… (truncated for display; full text preserved in the thread)");
	t
	} else {
	text.to_string()
	};
	let capped = if let Some((idx, _)) = text.char_indices().nth(MAX_BUBBLE_DISPLAY_CHARS) {
	let mut t = text[..idx].to_string();
	t.push_str("\n… (truncated for display; full text preserved in the thread)");
	t
	} else {
	text.to_string()
	};

[gemini-code-assist]

If the file does not exist or is unreadable, std::fs::metadata(path) will fail, and unwrap_or(0) will return 0. This causes the code to incorrectly report (image detected; will be sent as vision input) for a non-existent file. Handling the Result of metadata properly allows us to correctly report missing or unreadable files as skipped.

            if codescribe_core::attachment::image_media_type(path).is_some() {
                match std::fs::metadata(path) {
                    Ok(meta) => {
                        let size = meta.len();
                        if size > codescribe_core::attachment::MAX_VISION_IMAGE_BYTES {
                            out.push_str(&format!(
                                "(image too large for vision input: {} bytes > {} max; not sent)\n",
                                size,
                                codescribe_core::attachment::MAX_VISION_IMAGE_BYTES
                            ));
                        } else {
                            out.push_str("(image detected; will be sent as vision input)\n");
                            image_paths.push(display.to_string());
                        }
                    }
                    Err(_) => {
                        out.push_str("(skipped: image file not found or unreadable)\n");
                    }
                }
            } else {
                out.push_str("(skipped: unsupported image format or not UTF-8 text)\n");
            }

[Copilot]

Pull request overview

This PR improves the assistive voice-chat UX by (1) grouping and “de-wiring” tool evidence in the conversation timeline, (2) making image attachments reliably flow as real vision input with clearer user-visible error handling, and (3) improving Light Mode readability via deterministic contrast tokens; it also refines hotkey gating and adds an onboarding “Basic vs Agentic” lane with readiness probing.

Changes:

• Added grouped Tool Activity blocks (per assistant turn) and friendly tool labels/status messaging; tool failures now carry an explicit is_error bit to the UI.
• Centralized image-attachment marker parsing + vision loading in core::attachment, wired through both agent and legacy send paths, and added protection against empty image blocks.
• Introduced deterministic Light/Dark contrast palettes (WCAG-checked) and added onboarding lane persistence + agentic readiness probe/step flow.

Reviewed changes

Copilot reviewed 40 out of 40 changed files in this pull request and generated 8 comments.

Show a summary per file

File	Description
pico.save	Removed an editor scratch file from the repo.
docs/architecture/attachment-vision-pipeline.md	Added architecture doc for the attachment→vision pipeline and its shared marker contract.
core/tests/env_validation.rs	Tightened env-loader test isolation for llm_endpoint.
core/quality/qube_report.rs	Added protected-term loss detection to quality reporting.
core/pipeline/stream_postprocess.rs	Added operator vocabulary + protected terms lexicons; added lexicon re-apply and loss detection helpers; expanded tests.
core/llm/ai_formatting.rs	Reused shared attachment parsing/loading; increased image cap; applied lexicon post-LLM.
core/config/settings.rs	Added persisted onboarding_mode to settings (incl. env promotion + tests).
core/attachment.rs	Introduced shared image marker parsing + vision image loading + tests.
core/agent/session.rs	Tool-result summaries now surface first error message; propagates is_error to UI event.
core/agent/event.rs	Added is_error to AgentUiEvent::ToolResult for UI rendering.
assets/protected_terms.jsonl	Added curated protected-term lexicon entries (proper nouns/brands/tools).
assets/programming.jsonl	Updated Loctree canonical term/casing and expanded mispronunciations.
assets/operator_vocabulary.jsonl	Added Polish UI-command/operator normalization seed lexicon.
app/ui/voice_chat/tool_activity.rs	New pure module for per-turn grouped tool activity + evidence summaries + tests.
app/ui/voice_chat/state.rs	Added tool-activity group storage and “last sent attachments” state.
app/ui/voice_chat/mod.rs	Exported tool-activity APIs and module.
app/ui/voice_chat/handlers/menus.rs	Added “Re-attach previous” menu item; updated attachment menu logic.
app/ui/voice_chat/handlers/classes.rs	Registered new onAttachReattach: handler.
app/ui/voice_chat/handlers/attachments.rs	Implemented re-attach behavior for last-sent attachments.
app/ui/voice_chat/api/tests.rs	Added state-level tests for grouped tool activity rendering and turn boundaries.
app/ui/voice_chat/api/send.rs	Cleared attachments after send but preserved for re-attach; improved image honesty in attachment block.
app/ui/voice_chat/api/mod.rs	Imported tool-activity types used by the API layer.
app/ui/voice_chat/api/messages.rs	Added tool-activity record APIs; added bubble text capping/breaking to prevent CoreText hangs; added click toggling behavior.
app/ui/voice_chat/api/lifecycle.rs	Ensured tool-activity turn state resets on overlay clear.
app/ui/voice_chat/api/export.rs	Included Tool Activity role in markdown export labels.
app/ui/shared/helpers/mod.rs	Added deterministic Light/Dark contrast palette tokens + WCAG tests + appearance resolution.
app/ui/settings/engine_tab.rs	Added agentic readiness rows (gated by persisted onboarding mode); refactored tone→color mapping.
app/ui/onboarding/window.rs	Added onboarding lane chooser UI and readiness table UI scaffolding.
app/ui/onboarding/widgets.rs	Added mode radio syncing and a system orange color helper.
app/ui/onboarding/tests.rs	Added tests for onboarding mode persistence and lane-dependent step navigation.
app/ui/onboarding/steps.rs	Inserted Mode step; added AgenticReadiness step; kept stable index flow array.
app/ui/onboarding/state.rs	Added OnboardingModeChoice and initial mode selection from settings.
app/ui/onboarding/render.rs	Rendered Mode/AgenticReadiness steps; lane-aware sidebar markers; readiness table rendering.
app/ui/onboarding/handlers.rs	Added onModeSelected: handler wiring.
app/ui/onboarding/actions.rs	Added lane-aware step navigation; persisted onboarding mode; removed hardcoded full-disk step index.
app/controller/tests.rs	Updated hotkey gating tests; added predicate-level tests for assistive “Talk Anytime” vs raw start blocking.
app/controller/mod.rs	Allowed assistive start events during agent in-flight sends; kept Busy-state hotkey blocking.
app/controller/helpers.rs	Added friendly tool-name/status mapping; implemented image-attachment loading for agent send path; updated tool UI event handling; added tests.
app/agent/tools/mcp.rs	Added agentic readiness probe and tests; classified prerequisites including PRView detection heuristic.
app/agent/openai_provider.rs	Skipped empty image blocks when building provider input to avoid request rejection.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.