- Page number links with ellipsis for large ranges
- First/Last page buttons (« and »)
- "Go to page" input for direct page jumping
- Current page highlighted in indigo
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
When the Tools page loads, automatically sync statuses for all published
tools in the background. This means users don't need to manually click
"Sync Status" - the indicators will update automatically.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The publish dialog adds version and tags fields that aren't stored in
the local config.yaml. Exclude these from hash computation so the
hash comparison works correctly.
Also use Tool.to_dict() for consistent hash computation.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Track moderation status (pending/approved/rejected) in local tool config
- Add new "pending" state indicator (◐ yellow) distinct from "published" (✓ green)
- Add "Sync Status" button to check registry for updated moderation status
- Add /me/tools/<name>/status API endpoint for checking tool status
- Improve admin panel error handling with better error messages
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The tool list now refreshes after publishing to immediately show
the updated publish state indicator (✓).
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The Tool dataclass now has a path field that points to the config.yaml
file it was loaded from. This is set by load_tool() and enables the
publish dialog to save the registry_hash back to the local config.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add config_hash to publish API response
- Save registry_hash to local tool config after successful publish
- Show appropriate message based on moderation status (approved vs pending)
This enables the GUI to show publish state indicators (✓ published, ● modified)
for tools that have been published to the registry.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The publish_tool() API expects a YAML config string, but the dialog
was passing a dict directly, causing a 500 error on the server.
- Update PublishWorker to accept config_yaml string and readme
- Build complete tool config from Tool.to_dict()
- Convert to YAML string before sending
- Also include README.md if it exists in tool directory
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The _update_buttons method wasn't handling the no-token case properly.
When not connected, the Connect button should always be enabled.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add `cmdforge config disconnect` command to clear registry token
- Fix GUI connection status label not updating after connect
- Store status as instance variable instead of local
- Add _update_connection_status() method
- Call update after connect dialog completes
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Registry API improvements:
- Add /api/v1/me POST endpoint for profile updates
- Add /api/v1/me/password POST endpoint for password changes
- Fix visibility filtering on categories and tags endpoints to only
show approved public tools
- Allow underscores in tool names for fabric pattern compatibility
- Fix sqlite3.Row access (use bracket notation instead of .get())
Web UI improvements:
- Add tag filter to /tools page with three-state buttons
(include/exclude/neutral)
- Add mobile-friendly tag filter in responsive view
- Display tags on tool cards in the tools listing
- Add dashboard settings form handlers for profile and password
Admin improvements:
- Add scrutiny audit page for reviewing tool safety analysis
- Improve pending tools page with scrutiny report display
- Add scrutiny stats to admin dashboard
Fabric sync improvements:
- Add direct database publishing with scrutiny vetting
- Support auto-approve for tools passing scrutiny
- Improve error handling and logging
- Add source attribution for imported tools
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add hash_utils.py module for SHA256 content hashing with normalized YAML
- Store config_hash in registry database on publish
- Include hash in download response for client verification
- Verify downloaded content matches registry hash on install
- Store registry_hash in local tool config for publish state tracking
- Show publish state indicators in Tools page UI:
- Green checkmark: Published and up to date
- Orange dot: Modified since last publish
- No indicator: Local tool (never published)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add icons module with programmatically drawn icons:
- Speech bubble for prompt steps
- Code brackets </> for code steps
- Arrow icons for input/output nodes
- Display icons in list view next to step names
- Display icons in flow view node headers
- Add inline name editing in flow view via property_changed signal
- Sync name changes from flow view to tool model and list view
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Added "Cut links: Alt+Shift+drag" to help banner.
Shortened other labels to fit.
In NodeGraphQt, connections are cut by:
- Alt+Shift+Left-drag to slice through connections
- Dragging a connection endpoint to empty space
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add _rebuilding flag to ignore signals during rebuild
- Prevents cascade of port_disconnected signals when clearing graph
- Fixes KeyError in NodeGraphQt undo stack
- Fixes false dependency warnings on view switch
The port_disconnected and nodes_deleted signals were firing during
clear_session(), causing recursive rebuilds and errors.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
When a connection is broken in the flow view:
1. The node that lost its input moves to the end of the chain
2. The remaining nodes reconnect automatically
3. Tool model updates to reflect new order
4. Dependency validation warns about broken references
Flow: Input → A → B → C → Output
Break B's input: Input → A → C → B → Output
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Only parse {variable} patterns in PromptSteps, not CodeSteps.
In CodeSteps, variables are available directly as Python variables
(e.g., use `input` not `{input}`). The {var} syntax in code is
typically Python f-string or .format() syntax, not CmdForge
template substitution.
This fixes false positives like artifact-ai where Python f-strings
like f"{format_guide}" were incorrectly flagged as missing CmdForge
variable references.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
List View:
- Enable drag-drop reordering of steps
- Show broken dependency warnings when reordering
- Steps with missing variable refs shown in red with tooltip
Validation:
- Parse {variable} references from prompts and code
- Track available variables at each step position
- Warn when steps reference undefined variables
Flow View Sync:
- Deleting nodes in flow view updates the tool model
- steps_deleted signal propagates changes to builder
- Both views stay in sync
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- F key now fits to selected nodes (or all if none selected)
- A key selects all nodes
- Workflow: drag-select nodes then F to zoom, or A then F for all
- Updated help banner and context menu
- Context menu now has Select All (A) and Fit Selection (F)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Show banner on showEvent (when flow view becomes visible)
- Show banner on mouse enter event
- Raise banner to front to ensure visibility
- Recalculate size before positioning
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Remove Fit All button, F key shortcut is sufficient
- Add right-click context menu with "Fit All (F)" option
- Add floating help banner that appears on focus and auto-hides
- Banner stays visible when mouse hovers over it
- Fix default step naming to increment per type (Code 1, Prompt 1, Code 2)
instead of globally (Code 1, Prompt 2, Code 3)
- List view still shows overall step number (1, 2, 3) with per-type names
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Create welcome_page.py with branded landing page
- Add quick action cards for common tasks (Create Tool, Registry, etc.)
- Add "CmdForge" entry at top of sidebar navigation
- Update page indices for navigation
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add name field to PromptStep, CodeStep, and ToolStep dataclasses
- Update step dialogs to allow editing step names
- Display custom step names in both list and flow views
- Fix F key shortcut by installing event filter on graph widget
- Steps without custom names show default "Prompt N" / "Code N"
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add navigation help bar at top of flow view with instructions:
Pan, Zoom, Select, Edit step shortcuts
- Fix zoom to fit all nodes on initial load (was only centering last node)
- Use QTimer to ensure fit happens after widget is fully rendered
- Clear selection after fitting so nodes don't appear selected
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The setSelectionArea() API changed in PySide6 6.7 - it now requires
an ItemSelectionOperation parameter before ItemSelectionMode.
Files added:
- patches/nodegraphqt_pyside6_compat.patch: Git-style patch file
- patches/UPSTREAM_ISSUE.md: Issue report for upstream project
- scripts/patch_nodegraphqt.py: Script to auto-apply the patch
The patch fixes rubber band selection errors when dragging in the
node graph viewer. Run after pip install:
python scripts/patch_nodegraphqt.py
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Implements node-based flow visualization using NodeGraphQt-QuiltiX-fork:
New files:
- src/cmdforge/gui/widgets/flow_graph.py: FlowGraphWidget with custom node types
- InputNode: Shows tool inputs ($input + argument variables)
- PromptNode: AI prompt steps with provider display
- CodeNode: Python code steps
- OutputNode: Final output
- scripts/test_nodegraph.py: Standalone prototype for testing
Modified files:
- tool_builder_page.py: Added List/Flow view toggle with QStackedWidget
- Lazy-loads flow widget on first use
- Syncs between list and flow views
- Double-click nodes to edit steps
- styles.py: Added viewToggle button styling
- widgets/__init__.py: Export FlowGraphWidget
- pyproject.toml: Added 'flow' optional dependency group
Features:
- Toggle between List and Flow views in Steps panel
- Nodes auto-connected based on step order
- Auto-layout and fit-to-view on load
- Double-click nodes opens step edit dialog
- Graceful fallback if NodeGraphQt not installed
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Set default value "{response}" for new tools
- Add minimum height (80px) so it's always visible
- Add stretch factors to layout so Steps expands, Output stays fixed
- Update placeholder text to be more helpful
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Validates code with ast.parse() before accepting. Shows line number
and error message if syntax is invalid.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Feature was present in TUI but missing after GUI conversion. Now includes:
- Split layout: Code editor (left) | AI Assist panel (right)
- Provider selector dropdown for AI calls
- Prompt editor with smart default template showing available variables
- "Generate Code" button that calls AI in background thread
- Response feedback area showing success/error status
- Automatic markdown fence stripping from AI responses
- Available variables computed from tool arguments + previous step outputs
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add Profiles (AI Personas) section under GUI Features
- Add Ctrl+4 shortcut for Profiles page
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Profiles allow users to inject system instructions into prompts,
customizing the AI's behavior and persona for tool execution.
Features:
- Profile dataclass with name, description, system_prompt
- 8 built-in profiles: None, Comedian, Technical Writer, Teacher,
Concise, Creative, Code Reviewer, Analyst
- Custom profile creation and storage in ~/.cmdforge/profiles/
- Profile selector in Prompt Step dialog
- Profile injection during tool execution
- Profiles page in GUI (Ctrl+4) for viewing and managing profiles
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
search_tools requires a query, but list_tools allows browsing
without a search term. Use list_tools when no query or tags are set.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Browse all tools on page load without search
- Category filter dropdown (Text, Developer, Data, etc.)
- Sort options (downloads, rating, newest, name)
- Star ratings display in results table
- Clickable tags for filtering
- Installed indicator (✓) for local tools
- Update available indicator (↑) for newer versions
- Pagination controls for large result sets
- Publisher reputation info in details
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Web UI:
- Change sidebar "API Tokens" to "Connections"
- Change empty state button to "Make Your First Connection"
- Add TUI option to "About Connected Apps" info banner
- Add both TUI and CLI options to pairing modal
TUI:
- Change dialog title to "Connect to your CmdForge Account"
- Add account creation hint in username prompt
- Improve instructions with full URL and steps
- Fix timer refresh by calling loop.draw_screen()
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add pairing_requests table and hostname column to api_tokens
- Add pairing API endpoints: initiate, check, status, connected-apps
- Add cmdforge config connect <username> CLI command
- Rewrite tokens.html as Connected Apps with pairing flow
- Update TUI: Connect button when not authenticated, Publish when connected
- Add private sync option after save in TUI when connected
- Add visibility parameter to publish_tool in registry_client
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Was using werkzeug's generate_password_hash which creates incompatible
hashes. Now uses the same argon2 password_hasher as registration/login.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Fix _api_get to handle query strings in URL path
- Replace placeholder github.com/your-org URLs with gitea.brrd.tech/rob
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The index creation for moderation_status, visibility, role, and banned
columns was in SCHEMA_SQL which runs before migrate_db(). This caused
failures on existing databases that didn't have these columns yet.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Add documentation for running as a user service (systemctl --user)
which is how the current OMV deployment is configured. The previous
docs only showed system service setup which caused confusion.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
When the registry server starts, it now automatically adds any missing
columns to existing tables. This prevents 500 errors when publishing
to a database created before newer columns were added.
Columns added by migration:
- scrutiny_status, scrutiny_report (tool scrutiny)
- source, source_url, source_json (tool attribution)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
rob