Fix false text coverage by transparent areas and improve edit recovery
Distinguishes transparent areas from actual text coverage and verifies complete text in two viewports after edits in repeated real Gemma4 trials.
- Corrects visual checks so transparent navigation areas are not mistaken for text coverage.
- Provides visual evidence before long page text and retains current source and related styles during edit recovery.
- Confirms complete text on desktop and mobile after one edit in three fresh conversations with real Gemma4.
- Separates verified single-text repairs from other page warnings to make completion reports clearer.
Observe covered text and restore recheck evidence
Improves working-folder reuse and covered-text observations, and refreshes related style evidence during rechecks. Overall task completion remains based on actual request results.
- Reuses the verified path when setting or listing the same working folder again.
- Visual checks also flag text covered by another element, in addition to height limits.
- Refreshes related style evidence during rechecks to avoid repeating empty searches.
- Distinguishes missing model file actions from actual tool errors.
- Provides the model with observed positions and sizes of covering elements and bounds repeated rechecks without changes.
- Preserves the user-selected reasoning setting during recovery and aligns the displayed guidance with the actual request.
Verify unchanged text without unnecessary edits
Checks two viewports and the current file to report already visible text without unnecessary edits. Validation guards against treating mobile clipping or other edit requests as completed.
- Checks two viewports and the current file to report already visible text without unnecessary edits.
- Checks full text and observation limits, guarding against false completion when mobile clipping or other edits remain.
- Separates model response errors from visual checks and keeps overall developer tool readiness distinct.
Developer tool text verification and working folder display
Separates requested text checks from page-wide warnings and displays the current conversation working folder in the footer.
- Separates requested text visibility checks from page-wide warnings to reduce repeated edits to unrelated areas.
- Keeps the actual conversation working folder visible in the terminal footer and updates it after folder changes.
- When a lookup fails due to surrounding path whitespace, retries the same file listing once at the verified project path.
Developer tool edit completion and discovery scope
Checks execution evidence for edits without a filename and excludes dependency folders from default file listings.
- Checks actual edit evidence and verification even without a filename; unexecuted work is not treated as completed.
- Excludes dependency folders by default; query them explicitly or with the inclusion option when needed.
Developer tool text search recovery
Connects failed text searches to actual file paths and reduces initial reasoning waits for small repairs.
- Recovers failed searches containing border characters with a shorter literal lookup in the same scope.
- Reads surrounding source at actual search result paths to reduce guessed filenames.
- Uses a brief tool execution phase from initial exploration for small text repairs with default reasoning.
Developer tool exploration and edit recovery
Switches to focused repair after reading source and provides exact edit evidence for whitespace mismatches.
- Reduces unnecessary exploration history and starts focused repair once text and CSS source are available.
- Provides the unique exact source when an edit is rejected because of whitespace differences.
- Preserves existing files and reports why work remains incomplete when focused repair makes no progress.
Developer tool large-file navigation
The development source reduces repeated reads, connects search locations with relevant CSS evidence, and improves follow-up repairs.
- Reading the same range of a large file again does not resend unchanged source.
- Context around the searched text and relevant CSS in the same file are supplied as read evidence for the current hash.
- Automatic completion checks no longer block follow-up repairs after additional source inspection.
Developer tool PHP checks and path recovery
The development source improves PHP page checks and path recovery, with distinct runtime errors.
- A successful matching query at the verified candidate folder resolves a duplicate-path error.
- PHP text clipping repairs are checked on desktop and mobile through a temporary local server.
- Missing PHP runtime and page HTTP errors are distinguished when explaining incomplete checks.
Developer tool connection checks
The development source now follows an available credential check with a real server listing and distinguishes the connection result.
- For an explicit FTP/FTPS connection check, available credentials lead to a real listing on the same server. Checking credential status alone is never treated as a successful connection.
- Real connections keep the existing TLS policy and plain FTP consent. Credential-only or no-network requests do not connect.
- The tool records connection attempts separately from credential availability and reports observed transfer errors on failure.
Developer tool edit recovery
The development source now recovers missing edit arguments and detects text spilling on mobile.
- If an edit argument is missing, the original remains intact and recovery uses current code and its hash.
- Checks mobile text that spills beyond a constrained box and automatically verifies desktop and mobile after small HTML edits.
- Limits long reasoning and repeated calls during recovery, and no longer mistakes quoted instructional text for a how-to request.
Improved FONUMA AI response size display
Large response sizes now appear in MB during a task and in detailed status.
- The developer tool shows large live response sizes in MB, including detailed status.
Improved text clipping edits and verification recovery
Connects observed clipping with current source, then repairs and rechecks failed views.
- Finds current code for the affected element and its ancestors when text clipping is observed.
- Continues repairs using user constraints and current execution evidence after verification fails.
- Rechecks desktop and mobile after file changes and does not mark remaining clipping as complete.
Improved edit diagnostics and text clipping completion checks
Clarifies partial reads and source mismatches, and identifies remaining text clipping during completion checks.
- Shows returned line ranges and the next read position for partial file reads, with specific source mismatch guidance.
- Recognizes edit requests with Korean particles after filenames and checks once before ending without a change.
- Marks browser verification incomplete when potential text clipping remains after a requested clipping repair is saved.
Thinking controls and text clipping observations
Choose thinking controls and inspect potentially clipped text and elements in the browser.
- Request thinking on or off even when the server omits its controls, with support shown as unverified.
- Browser checks include potentially clipped text and its element and ancestor information.
- Separate changing task state from fixed instructions and record input processing, generation, and loading times in model diagnostics.
Distinguishing recovered server connection checks
Separates successful connection checks from access failures for other remote folders and provides clear next steps.
- For connection-only requests, a successful listing on the same server confirms connectivity while other folder access failures remain as notes.
- Failures for a specific remote folder or download remain unresolved even when another folder succeeds.
- Identifies remote directory access refusal separately and removes unnecessary blanket warnings when no unresolved errors remain.
Corrected repeated project names in working paths
Resolves relative paths with a repeated project name to a verified internal folder and shows the applied location.
- When setting a working directory, corrects repeated project names in relative paths to a verified internal folder and shows the applied path.
- Uses the original path when it exists, without creating missing folders or bypassing link and protected-path checks.
Improved connection checks and repeated exploration
Separates credential configuration from connection checks and refines repeated exploration and reasoning-only output budgets for small layout fixes.
- Distinguishes credential checks from actual server connections and corrects completion replies that lack connection evidence.
- Reassesses repeated exploration during small clipping fixes and explains incomplete work when there is no progress.
- Does not automatically increase the output budget when a limit is reached with reasoning alone.
Light theme panel and input colors
Refines light-theme accents, search and run selectors, and the add button background while preserving dark-theme colors.
- Uses pure red in the left panels and replaces green accents with blue in the light theme.
- Matches global-search inputs and selectors and the run selector with gray backgrounds.
- Gives the launcher add button a white background and keeps existing dark-theme colors.
Search context and server error guidance
Provides surrounding code and file hashes in search results and clarifies what AI server errors establish.
- Search results include surrounding code and file hashes so small edits can proceed without a separate file read.
- Marks omitted or truncated search context and uses hash checks to reject edits when a file has changed since the search.
- Explains AI server responses 502, 503 and 504 and guides checks of server status and existing work.
Status checks and task completion
Checks whether edits were saved without extra work and clearly reports verification and runtime requirements.
- Standalone edit-status questions compare saved records with current files without further edits.
- Improves completion handling and repeated-error guidance so checks can finish without changing files.
- Final explanations distinguish saved files, execution and visual checks, and server deployment.
- Explains runtime requirements before previewing files such as PHP and avoids flagging new code evidence as repetitive exploration.
Focused search and clear cancellation results
Focuses searches on located files and clarifies repeated exploration and cancelled downloads.
- Searches directly within located files and excludes external libraries by default to focus on project files.
- Explains duplicated relative paths and invalid search cursors, and detects repeated searches.
- Shows change status and guidance to narrow the scope when file exploration continues.
- Distinguishes cancelled downloads from transfer failures while preserving completed files and existing failure records.
Transfer progress log colors
Download status appears in gray with the same font size.
- Download and transfer progress logs use gray text at the same size as the main text.
Server transfers and task recovery
A shared executor handles folder downloads and resumption, distinguishing request cleanup from work that has not started.
- Downloads server folders recursively and records progress, failures, and exclusions.
- Verifies size, modification time, and hashes to reuse completed downloads while preserving existing files.
- Improves directory paths and password corrections, blocking calls to an unintended server.
- Rechecks cleanup of the previous model request and explains blocked requests and unknown states.
- Records actual generation submissions without mislabeling a pre-request block as a file save failure.
- Synchronizes terminal dimensions with completed frames so the input starts on the correct row.
Spacing below AI replies
Added a blank line between AI replies and task logs.
- Added one blank line between AI replies and the task log below.
Failure explanations and follow-up questions
The development build explains failed steps and unknowns, then asks for the information needed to continue.
- Connection and authentication failures distinguish facts from unknowns and ask for information needed to resolve them.
- An error explanation and next question appear even when the model returns an empty final answer.
- The tool waits for the next user request without repeating the same blocked work.
Conversation cursor and terminal text improvements
The development build preserves the input cursor after replies and matches terminal text to the selectable transcript.
- Draft text and cursor position remain intact when an AI reply ends.
- Dark terminal text now uses the same bright color as the selectable transcript.
Credential saving and test-app shutdown improvements
The development build improves explicit credential saving, original input display and draft confirmation.
- Explicit save requests with credentials are handled directly by secure storage within granted permissions.
- User-entered passwords appear as typed in the terminal; unavailable passwords are not claimed to be recoverable.
- Draft read failures are separated from write failures; test apps close after saving is confirmed.
Connect and save server credentials in the terminal
The development build lets you choose plain FTP and credential saving without leaving the terminal.
- Choose whether to use plain FTP in the terminal and continue the same task.
- After connecting, choose secure saving or use for this request only.
- Reuse saved server credentials and connection settings in later requests.
Clearer guidance after connection failures
The development build explains what a TLS failure establishes and which settings to check next.
- TLS failures show login and listing as unverified and provide next steps in server authentication settings.
- Connection guidance remains even when the model reply is empty, with fewer model repair calls for TLS configuration errors.
Server authentication and one-time connections
The development build supports dedicated credential entry, secure storage and request-only authentication.
- Check a connection once or register it in OS secure storage from the server authentication screen.
- Explicitly supplied connection credentials are used only for that request and masked in model input and chat history.
- Plain FTP requires user selection, and registration is reported separately from actual connection results.
Connection setup and empty-response recovery
The development build checks local authentication configuration and reports outcomes after empty error-recovery replies.
- Checks fixed local authentication configuration without exposing file contents.
- Blocks transfers to hosts confirmed to lack local authentication configuration and provides setup guidance.
- Shows observed results and unresolved errors even if the recovery reply is empty.
Clearer outcomes and validation scope
The development build separates target-specific errors and success evidence and documents workflow validation.
- Successful operations and unresolved errors appear together, scoped to each target.
- Connection checks no longer acquire an unrelated file-save requirement.
- TLS negotiation and login failures are distinguished; another credential input format is masked.
- Documentation distinguishes automated checks, isolated execution and production-server verification.
Improved scrolling after developer tool exit
Retained conversations can be scrolled after forced exit in the development build.
- Received conversations are restored to scrollback after forced exit.
- Reopening saved records restores the retained conversation.
- Missing conversation data is reported as a recovery limitation.
Improved command recovery and server file work
The development build improves command recovery, credential protection, multi-file reads and TLS transfers.
- Require help inspection after option errors and retain failure counts when arguments change.
- Mask recognized credentials in model input, display and saved records, and block commands containing them.
- Read related files together with per-file hashes and partial-read notices.
- TLS listings and new-file downloads include size and hash checks, existing-file protection and cancellation cleanup.
- The transcript view wraps long lines without horizontal scrolling.
Improved file destinations and collision recovery
The development build checks new file destinations early and reduces content regeneration after existing-file conflicts.
- Supported single-file creation requests check the destination and existing files in the current project before generating content.
- Existing-file conflicts preserve the original and retain generated content temporarily so only the destination needs correction.
- Limit destination repair attempts and count repeated file conflicts regardless of content changes.
Improved requirement checks and terminal readability
The development build separates save, layout and reference checks and brightens dark-mode text.
- Show save, layout and reference comparison results separately from unverified requirements.
- Compare supported company, contact and service references at two screen widths and request focused edits for mismatches.
- Block whole-file replacement, deletion and moves during verification repair to preserve the saved artifact.
- Slightly brighten the default terminal text in dark mode.
Improved CLI history scrolling
The development build improves vertical scrollbar visibility and browsing earlier output in regular Codex sessions.
- Terminals with history keep the vertical scrollbar visible, making it easier to find and drag in dark and light modes.
- Regular Codex launches use the supported history-preserving mode so earlier output can be browsed with the wheel and scrollbar.
Improved static website verification
The development build reduces unnecessary test commands and repeated work caused by configuration errors.
- New projects no longer prefill a test command, avoiding unnecessary npm checks for static HTML tasks.
- npm check failures identified as project configuration errors no longer trigger file rewrites or repeated model reasoning.
- Project test results and browser checks are reported separately, and available page checks continue.
Fixed file-path request input
The development build now distinguishes file-path requests from menu commands.
- Fixed task requests starting with an absolute file path being mistaken for menu commands.
- Multiline file requests pasted during work queue as one request and preserve their original text.
More reliable development tool search and verification
Improved search, logs, project instructions, final checks, and task continuity in the development build.
- Continue through file lists and search results in large projects, with unsearched areas identified.
- Keep showing recent progress logs during large command output and preserve Korean text.
- Load project instructions automatically and review newly discovered instructions before changing files.
- Run configured project tests after the final edits and follow failures with repair and rechecking.
- Publish complete new files in one step and preserve existing files on conflicts.
- Retain bounded excerpts of the first request and recent corrections after summarization to support continuity.
Artifact Check: saved files and browser checks
Improved task completion based on saved results and HTML browser checks.
- File task completion messages now use actual save results.
- After saving HTML, desktop and mobile browser checks identify observed issues.
- Combining the final save and finish request avoids waiting for an extra closing response.
- Failed checks or external edits preserve files and are reported as incomplete verification.
Terminal Spacing: restored terminal padding
Restored the terminal’s original right padding.
- Restored 10px of right padding to match the terminal’s left spacing.
Live Progress: activity and permissions
Improved live response and code visibility, file permissions, and terminal appearance.
- Track work through actual response reception and unsaved code previews.
- Wait limit notices move to status details while command output remains visible.
- With full access selected, create ordinary files outside the project and repair missing destinations.
- Dark terminal text matches sidebar descriptions, with less unused space on the right.
Tool Recovery: file recovery and light theme readability
Improved recovery from missing file paths and text readability on light backgrounds.
- When a file destination is missing, reuse the retained body and request only the missing path.
- Repeated tool argument errors stop retries and clearly mark the task as incomplete.
- Light mode uses pale gray backgrounds for the left panels and higher contrast for terminal text.
Light Theme: dark and light appearance
Switch between light and dark while working and keep your preferred appearance.
- Choose a dark or light theme in Settings or the View menu.
- Terminal backgrounds, input areas, and developer tool conversations and menus change together.
- Keep running work and drafts intact, with your selected theme restored when the app reopens.
Task Report: work records and PDF creation
Improved task timing records and Korean text PDF creation in developer tools.
- View measured elapsed time, model wait time and tool call counts for each request.
- Preserves executed results and displays task records even when the final reply is empty.
- Creates Korean text PDFs without installing additional programs and preserves existing files.
Scrollback: browse output after exit
Improved conversation retention and terminal history scrolling after exit.
- After a normal exit, retained development tool conversations remain available for scrolling.
- Released mouse reporting in exited terminal history to restore wheel scrolling and text selection.
Request Status: server state and generation flow
Improved server status and cancellation checks, and reduced repeated file-generation work.
- Reference files use available context, and automatic output capacity increases ahead of generation based on file size.
- Preserves public response text so generation-limit retries can retain the stated design and scope.
- Shows per-request server status and actual time limits, separating connection checks from observed output.
- Esc cancellation checks server connection cleanup and slot release, with engine termination observability shown separately.
- Prevents new generation while cleanup of a previous server request remains unconfirmed.
Long Work: response waits and file replacement
Improved waits for long tool responses and replacement of existing files.
- Removed the separate five-minute HTTP limit on model connections so the development tool controls waiting and cancellation.
- Waits for long tool responses adapt to measured generation speed and output size; completion no longer waits for the connection to close.
- Requests that time out without confirmed server cancellation are no longer retried automatically.
- Added a tool to replace existing files with complete content without deleting them first, with cancellation, conflict and undo checks.
- Verbose build and test logs no longer terminate commands; output tails and actual exit codes are preserved.
Output Budget: automatic generation budget
Improved automatic generation budgets and limit notifications.
- Removed output budget reductions based only on generation speed.
- Automatic mode increases the output budget within available context when generation hits its limit; explicit user caps are preserved.
- Limit notifications distinguish reasoning-only output from incomplete answers or tool requests.
Loop Recovery: task progress and response recovery
Improved stalled-response recovery and task progress and verification records.
- If a model response stalls, keep completed tool results and recover the unfinished request once.
- Adjust the next response output budget using the server's measured generation speed.
- Track whether checks followed the latest changes, flag repeated work, and do not mark empty responses complete.
Input Controls: input and task switching
Improved Esc cancellation, multiline paste and queued request switching.
- Press Esc in the task view to cancel the current AI task and queued requests.
- Multiline paste stays in one draft until you press Enter to submit.
- If the server returns HTTP 429 after switching requests, wait briefly and retry a limited number of times.
Task Efficiency: development tool efficiency
Reduced repeated reads and premature context summaries, and improved local previews and server identification.
- File changes return the latest hash to reduce unnecessary reads before the next edit.
- Context budgets use server-reported input token counts to reduce premature summaries.
- Fixed relative paths in local HTML previews and preserve full development server IDs.
Immediate Menus: commands while waiting
Function menus now open immediately while waiting for an AI response.
- Open the / command list and permissions menu immediately while waiting for AI.
- Commands bypass the prompt queue; additional questions still support queuing, editing and deletion.
File Results: verify file operations
Improved file-write failure messages and execution result checks.
- Withholds unverified model responses while tool errors remain and rechecks execution results.
- Shows the path and content length first in file-write logs.
- Distinguishes parent traversal, protected paths and missing files or folders.
Work Indicator: activity animation
The hexagon icon softly fades while the model is working.
- The hexagon above the input softly pulses during model reasoning and response generation.
- Preserves the accent color and shows a steady icon when reduced motion is enabled.
Reference Path: verify reference file locations
Preserves specified file addresses and detects same-named references at different locations.
- Converts file:// URLs to filesystem paths to preserve reference locations.
- Reports a path mismatch when a same-named file differs from the specified reference location.
- Shows the resolved absolute path in read results and uses existing permissions for references outside the project.
Queue Controls: manage queued requests
Added menus during tasks and queued-request controls, with stable input rendering.
- Open the / menu and choose commands with arrow keys while AI is responding.
- Commands selected during a task are queued; stop and exit take effect immediately.
- Use the menu on the right of a queued request to run it now, edit it or delete it.
- Keeps the input position and black background stable during scrolling and redraws to prevent overlapping content.
Chat Icons: distinguish speakers
Distinct user and AI icons with white response text.
- Marks user questions with the original small › icon.
- Shows AI responses with a filled accent-color hexagon ⬢ and white text.
White Reply: white AI responses
Changed AI response text to white.
- Developer tools display AI response text in white.
Language Menu: developer tool languages
Added multilingual menus, compact context values and conversation icons.
- Choose Korean, English, Japanese or Vietnamese in the developer tool /language menu.
- Applies the chosen language to menus and guidance and keeps it for future launches.
- Context status and welcome information also use compact lengths such as 128K and 256K.
- Uses ▣ for AI responses and ☻ for user questions while keeping wrapped text aligned.
History Selection: delete selected runs
Delete multiple selected runs and view concise context information in the footer.
- Select multiple entries or select all in the run-history deletion dialog.
- Selected active runs are stopped before deletion, with completed counts shown.
- Footer limits use compact values such as 32K and 256K, labeled as the configured limit.
- Shows the model name first in footer information.
- New conversations and the context menu now default to 128K.
Large Context: expanded context presets
Added larger context options and separated default selection from active settings.
- Added 512K and 1024K presets, displayed according to the reported model limit.
- The menu defaults to 32K and marks the currently configured value as active.
Context Presets: context selection menu
Select context length from a menu, with a highlighted copy-view return button.
- Select and apply 4K, 8K, 16K, 32K, 64K, 128K or 256K context with arrow keys.
- Shows current settings and server/model limits while retaining direct numeric input.
- Highlighted the copy view return button with the accent color.
Model Context: inspect context limits
Compare tool settings, server context and model capacity.
- Below thinking on the welcome screen, separately shows tool, server and maximum model context.
- Context details and the footer also show server values; unavailable values are marked unknown.
Approval Menu: task approval choices
Select task approvals by keyboard and apply them immediately.
- Approval requests appear above the input, with arrow keys and Enter to approve, deny or stop.
- Approving all actions in the current task applies immediately after selecting the item that describes its scope.
Welcome Accent: highlighted welcome
Refined the opening prompt and guide divider.
- Highlights the welcome prompt in the accent color and adds a divider below temporary chat guidance.
Composer Info: input layout and usage display
Improved initial input layout, settings information and context/token display.
- Applies the current input design from first connection and reduces the gap below logs.
- Groups conversation and connection settings in a shaded block with distinct label colors.
- Shows estimated context, configured limit and recent input/output tokens to the left of the model name.
Reply Colors: distinct AI response colors
AI replies and completion notices use distinct colors.
- AI replies use the accent color and completion notices use blue.
Permission Choice: apply selections immediately
Removed repeated confirmation input after permission selection.
- Selecting approve all in the permissions menu now applies immediately without typing yes again.
Composer Layout: aligned input and menus
Refined input centering, menu spacing, model display and progress log colors.
- Centered input text vertically between smaller top and bottom padding rows.
- Placed menus and progress just above the input, with refined footer spacing and indentation.
- Progress and elapsed time beside the accent hexagon are shown in white.
- Shows the selected model below the input in the accent color and updates it when the model changes.
Work Header: progress above the input
Moved the accent-colored hexagon icon and task progress above the input.
- Placed an accent-colored hexagon, progress log and task time in one line above the input.
Composer Footer: less empty space below
Placed the input hints at the bottom to reduce unused space.
- Aligned input hints and status text to the bottom edge to reduce the empty space beneath them.
Composer Compact: input spacing and cursor
Reduced footer text and spacing, and changed the input cursor to an underline.
- Reduced spacing below the input and set hint and status text to 10px.
- Reduced inner side padding to two columns and changed the cursor to a blinking underline.
Welcome Labels: clearer startup details
Startup labels and values are now distinct and aligned.
- Startup labels are bold, with description, developer, website, model, reasoning and working folder values aligned.
Composer Borderless: removed input borders
The conversation input is now a borderless black area.
- Removed the conversation input border while retaining its black background and inner padding.
FONUMA Conversation: conversation display and context management
Developer tools dev.9 improve the composer, queued request switching, compact execution logs and copy selection. Verified with the real app and a model fixture; public installers are unchanged.
- Added a black composer with side padding and rounded borders; task time now appears below it on the right.
- Queued requests stay visible one per line. Press Enter with empty input to stop the current task and run the first queued request.
- Execution logs show tool names and action types, with short previews of long results.
- Added bold reply text, a vertical scroll position indicator and a copy view that preserves selection while scrolling.
- Large recent tool exchanges are summarized further to make room for replies. Error text inside code no longer clears the working indicator.
FONUMA Generation: generation limits and queued prompts
Developer tools dev.8 add generation-limit recovery, queued prompts and elapsed task time. Verified with a model fixture and the real app; public installers are unchanged.
- Raised the default generation budget to 8,192 tokens. Truncated tool calls are not executed; one smaller step is requested instead.
- Additional prompts wait in order until the current task completes; cancellation or failure prevents automatic execution.
- Shows elapsed time since the prompt started at the upper right of the input area.
Composer Status: work indicator positioning
Moved the working indicator to the upper right of the input area. Verified in the real app; public installers are unchanged.
- Shows the working indicator in small gray text at the upper right of the chat input, hidden while waiting for input.
- Reduced the empty bottom row around additional input during work for a more compact layout.
AI Switch: switch dialog and run history layout
Improved the AI switch dialog and run history selector. Verified layout, switching and cancellation in the real app; public installers are unchanged.
- Organized the switch dialog description, selection, confirmation and buttons to fit narrower windows.
- Improved AI list loading and confirmation after selection changes; verified cancellation and launching the selected AI.
- Reduced the run history selector width to leave more space for adjacent controls.
- Fixed AI switching stopping after the old run ended without starting the next AI.
FONUMA Long Work: optional execution limits
Developer tools dev.7 make time and request limits optional. Verified with simulated long responses, real commands and the app; public installers are unchanged.
- Removed default task time and request limits and enabled longer commands and test runs.
- Models with continuing responses are not stopped by elapsed time; cancellation, stalled responses and repeated-error detection remain.
FONUMA Code: code display, progress and selection
Developer tools dev.6 improve code display, progress, text selection and model waiting. Verified with the real app and a model fixture; public installers are unchanged.
- Code uses a black background and syntax colors, with red and green lines for before and after edits.
- Diagnostic logs appear in gray and task progress updates on one line.
- Improved Shift-drag selection after scrolling and preserved selection during progress updates.
- Model data waiting now allows five minutes and indicates when incoming data has stalled.
FONUMA Welcome: startup screen improvements
Improved the English heading and product introduction on the developer tools welcome screen. Applied to development sources; public installers are unchanged.
- The welcome screen displays a large FONUMA banner and the FONUMA DEVELOPER TOOLS heading.
- An introduction, developer name and website appear below the version.
FONUMA Work Memory: long tasks and conversation layout
Developer tools v0.1.0-dev.5 add structured work summaries, resuming, scrolling and input readability. Verified with deterministic model fixtures and the real app; public installers are unchanged.
- Added structured work memory for goals, decisions, verification and remaining work, with /memory inspection.
- Resuming a saved conversation restores its summary, next steps and context setting; failed summaries preserve the original conversation.
- Improved mouse-wheel scrolling, vertical input padding, question backgrounds and hanging indentation for wrapped replies.
AI Connection: connection mode in registered AI
Registered AI now shows its connection mode and endpoint in development. Verified in the app; public installers are unchanged.
- Registered AI cards show remote HTTPS, loopback Ollama/SSH tunnel or CLI-managed connection settings alongside the endpoint.
FONUMA Model: clearer connection errors
Developer tools v0.1.0-dev.4 distinguish model-list connection failures. Verified in development; public installers are unchanged.
- The model menu distinguishes connection failures, timeouts and invalid responses, with guidance for local ports and SSH tunnels.
FONUMA Agent Chat: conversation and menu improvements
Developer tools v0.1.0-dev.3 add fixed input, keyboard menus and status colors. Verified in development; public installers are unchanged.
- Redesigned the welcome screen and fixed a distinct input area at the bottom.
- Type / first to open the menu immediately; use arrows, Enter and Esc to navigate menus and submenus.
- Added menu backgrounds, bold headings, green selections and status-specific log colors.
AI settings: connection form position
In development source, Add AI connection appears above Registered AI. Verified in Electron; not yet included in public installers.
- Moved the connection form above registered AI and updated the empty-list guidance and divider in development settings.
Ollama HTTPS · remote server authentication (development source)
Register a remote Ollama server address and authentication method for developer tools. Verified in development; not publicly released.
- Added remote HTTPS addresses with no authentication, username/password or API key to Ollama connections.
- Credentials are kept in macOS Keychain; model lookup and FONUMA/Codex developer tools share the same connection.
- Preserves existing local/SSH connections and displays FONUMA developer tools v0.1.0-dev.2.
FONUMA Agent UI · tool name and version (development source)
Registered AI connections show the developer tool name and version together. Verified in the development UI; not publicly released.
- Shows the tool name and version together in the developer tool version field of registered AI connections.
FONUMA Agent version display (development source)
View the actual FONUMA developer runtime version in registered AI connections and terminals. Existing model and port registrations are preserved; no public release was made.
- Shows the developer tool version separately from the model name in registered FONUMA AI connections.
- The terminal startup message and /status also show the connected runtime version.
- Older runtimes without version metadata show a restart notice while keeping AI registrations unchanged.
FONUMA Preview · local pages and development servers (development source)
Local page inspection and development-server lifecycle are connected to FONUMA developer tools. Verified in development source and a local test app; not publicly released.
- FONUMA developer tools can inspect local HTML and pages from development servers they start.
- Inspect desktop/mobile DOM, console errors, horizontal overflow and temporary screenshots.
- Adds development-server approval, readiness checks, logs and stopping; owned servers are cleaned up when the task ends.
- Restricts preview access to external sites, other local ports, file links and downloads, isolated from existing app sign-in data.
Unified FONUMA developer runtime (development source)
Conversation and file/command work now share one session, with improved approvals, cancellation and history. This is a development-source change, not a public installer release or completion of the full final design.
- FONUMA developer tools keep chat and file/command results in one conversation without requiring /work.
- Adds file-change approval, protection against conflicting user edits, undo for direct file changes and Ctrl+C cancellation.
- Explicitly saved conversations resume by exact ID; long tool results can be paged without rerunning commands.
- Configure model, context and execution budgets in the CLI; repeated results and unresolved tool errors produce explicit failure notices.
- Connects native execution and approval status to terminal lists and tabs, and adds a standalone Mac terminal entry point.
Beta 116 restore progress, AI status, and command launcher import
Restore progress is now clearly shown, list and tab status indicators work with the latest Codex CLI, and executable command files can be safely imported as launchers.
- During restore or resynchronization, a spinner and input-lock notice appear over the terminal, then the latest screen is shown when complete.
- Codex CLI 0.155.1 AI connection and work states now restore the status colors and work icons in the left list and terminal tabs.
- Launcher import accepts FONUMA JSON and executable .command, .sh, .zsh, and .bash files for review before registration.
- Importing a command file does not execute or read its contents; symbolic links, non-executable files, and unsupported formats are rejected without changing the original.
