Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

- v2-phase5.AC11.2: Add test_first_attempt_has_no_previous_attempt
Verifies that previous_attempt is None on first attempt. Creates a task
with no "previous_attempt" key in metadata and confirms the orchestrator
correctly derives previous_attempt = None when building AgentContext
(orchestrator.rs:793-795).

- v2-phase5.AC12.2: Add test_unrelated_task_unaffected_by_failure_cascade
Verifies that an unrelated task C (no dependency on failed task A) remains
in Ready status when failure cascading occurs. Tests that cascade blocking
is dependency-aware and does not affect unrelated tasks.

All 467 tests pass.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Reflect new autonomy module, CodeSearchTool, budget-aware context builder,
orchestrator error recovery, and SecurityScope enforcement methods.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Addresses two code review issues from Phase 6:

Critical Issue 1: Integration test did not exercise try_unblock_tasks
- Location: tests/orchestrator_test.rs in test_retry_cascade_unblock_lifecycle
- Root Cause: Test manually reimplemented the unblock logic using direct graph_store
calls instead of calling orchestrator.handle_scheduling() which calls
try_unblock_tasks in production code
- Fix Applied: Replaced manual unblock logic (lines 838-862) with call to
orchestrator.handle_scheduling().await. This ensures the actual production
code path (try_unblock_tasks) has test coverage.
- Verification: Task B status transitions from Blocked to Ready/Claimed after
handle_scheduling executes try_unblock_tasks, confirming the production
path was exercised

Important Issue 1: Missing test for AC13.2 (Blocked task stays Blocked when
blocker is still Failed)
- Location: tests/orchestrator_test.rs in test_retry_cascade_unblock_lifecycle
- Root Cause: Test skipped from verifying cascade (B blocked) directly to
completing A, missing the intermediate verification
- Fix Applied: Added Step 3 that verifies task A remains Failed and task B
remains Blocked before completing A. This confirms AC13.2.
- Verification: Assertions verify both tasks maintain their states before
the unblock operation

Additional fix:
- Fixed formatting issue in src/tools/search.rs line 333 (cargo fmt)

All tests pass: 557+ tests including full orchestrator_test suite

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

- Critical Issue: Filter node_modules and target directories in code search.
The filter_entry closure now checks skip_dirs for ALL directories, not just
those starting with '.'. This ensures node_modules and target are properly
skipped. Added unit test code_search_skips_node_modules_and_target to verify.

- Important Issue: Collapse nested if-let chains in orchestrator.rs using
Rust 2024 let-chains. The try_unblock_tasks method now uses a single
if let ... && let ... condition instead of nested blocks, resolving the
clippy collapsible_if warning.

- Minor Issue: Tighten assertion in code_search_skips_binary_files test.
Removed the vacuous disjunction - the '.' regex always matches text.txt,
so the assertion now directly checks for "text.txt" presence.

Verification:
- All 7 code_search unit tests pass
- Full test suite passes (all 100+ tests)
- cargo clippy clean (collapsible_if warning resolved)
- cargo build successful

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Phase 5 - Agent Error Recovery & Task Reassignment (Tasks 1-4)

Implements three critical orchestrator enhancements:

1. Previous Attempt Context Injection (AC11.1, AC11.2):
- When a task is retried, the previous failure error is stored in metadata['previous_attempt']
- This error is injected into the retried worker's AgentContext so it can learn from the failure
- On first attempt, previous_attempt is None

2. Failure Cascading to Dependents (AC12.1, AC12.2):
- When a task fails permanently (retries exhausted), cascade_block_to_dependents() finds all
downstream tasks linked via DependsOn edges and marks them Blocked
- Stores blocker task ID in metadata['blocker_task_id'] for reliable unblock tracking
- Non-dependent tasks remain unaffected

3. Blocked Task Recovery (AC13.1, AC13.2):
- During scheduling, try_unblock_tasks() checks all Blocked tasks
- Tasks whose blocker has been completed are transitioned back to Ready
- Tasks whose blocker is still Failed/Blocked remain Blocked
- Uses metadata['blocker_task_id'] for reliable lookup (no string parsing)

4. Integration Test Coverage:
- test_retry_cascade_unblock_lifecycle exercises full lifecycle from retry through
cascade to unblock, verifying all acceptance criteria

All 23 orchestrator tests pass, full test suite (405 tests) passes.
Code formatted and linted without warnings.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Replace closure with Pattern::new function reference per clippy suggestion.

Implements Task 3 from Phase 5: Add comprehensive integration tests verifying
code search functionality including pattern matching, file glob filtering,
result capping, directory scoping, and tool registration in the V2 registry.

Implements Task 2 from Phase 5: Add CodeSearchTool to create_v2_registry with
project_root parameter. Update all call sites (orchestrator and test files)
to pass project_root path.

Implements Task 1 from Phase 5: Create CodeSearchTool that searches file contents
with regex pattern matching, optional file glob filtering, directory scoping, and
result capping. The tool skips binary files and respects the project root boundary.

Auto-formatted format! macro on line 85 to fit within line length constraints.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Fix 3 clippy warnings for collapsible_if statements in check_path and
check_command methods. Use let-chains (stable in edition 2024, Rust 1.85.0+)
instead of nested if let/if blocks for more idiomatic code.

- Lines 82-89: check_path denied patterns loop
- Lines 94-99: check_path allowed patterns loop
- Lines 107-111: check_command allowed commands loop

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Apply rustfmt formatting to src/security/scope.rs and tests/security_scope_test.rs
to match project conventions.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Export the new FileOperation and ScopeCheck enums from the security module
so they can be used in integration tests.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Add comprehensive integration tests verifying that built-in profiles (planner,
coder, reviewer, tester, researcher) correctly enforce their security scopes.
Tests verify read-only constraints, file creation restrictions, command patterns,
and network access controls across all profiles.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Add FileOperation and ScopeCheck enums to enable enforcement of path and command
validation. Implement check_path with support for glob patterns, read-only mode,
and file creation restrictions. Implement check_command for shell command validation
and check_network for network access control. Include comprehensive unit tests
covering all acceptance criteria.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

- Replace manual Default impl with #[derive(Default)] on AutonomyLevel enum
and add #[default] attribute to Supervised variant. This resolves the
clippy warning about derivable_impls.
- Fix formatting in test code (lines 378, 390, 402) via cargo fmt.

These changes resolve Important Issue 1 and Minor Issue 1 from code review.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Implement autonomy type system for approval gates with three levels:
- Full: No approval gates active
- Supervised: PlanReview, PreCommit, TaskComplete, GoalComplete gates active
- Gated: All 6 gates active (adds DecisionPoint, WorkerSpawn)

Default level is Supervised. Includes helper methods for checking
active gates and is_gate_active for individual gate validation.

Verifies v2-phase5.AC5.1, AC5.2, AC5.3, AC5.4

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Important Issue 1: Formatting
- Fixed import order in src/agent/runtime.rs:2 (ContextBudget, ContextBuilder)
- Ran cargo fmt to fix line-length and struct literal formatting issues

Minor Issue 1: Test Mock Duplication
- Extracted TestGraphStore trait mock (~90 lines) to module level (src/context/mod.rs:356-447)
- Created create_test_context() helper function to reduce duplication across 6 test functions
- Refactored test functions to use shared helper and removed ~540 lines of duplication:
- test_build_system_prompt_output_format
- test_v2_phase5_ac3_1_dependency_status_done
- test_v2_phase5_ac3_2_previous_attempt_present
- test_v2_phase5_ac3_3_previous_attempt_absent
- test_v2_phase5_ac4_1_budget_aware_trimming
- test_v2_phase5_ac4_2_required_sections_never_trimmed

Verification:
- cargo fmt --check: passed
- cargo test --lib: 62 tests passed
- cargo clippy: no new warnings
- cargo build: success

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Update AgentRuntime::run to use build_system_prompt_with_budget with default
ContextBudget instead of the unbounded build_system_prompt. This ensures that
the system prompt is constructed with token budget awareness.

Verifies v2-phase5.AC4.1

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Implement ContextBudget struct and build_system_prompt_with_budget method to manage
priority-based token allocation and overflow trimming. Required sections (Role, Task,
Dependencies, Previous Attempt, Rules) are never trimmed. Optional sections (Session
Continuity, Active Decisions, Observations, Project Conventions) are trimmed from
lowest priority up when over budget.

Verifies v2-phase5.AC4.1 and v2-phase5.AC4.2

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Implements v2-phase5.AC3.1, AC3.2, AC3.3:
- Add dependency status lines with [DEP:DONE] and [DEP:PENDING] markers
- Add Previous Attempt section with [PREV_ATTEMPT] marker when retry attempt exists
- Omit Previous Attempt section when no previous attempt is present

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

- Minor Issue 1: Update parameter description for path field to indicate
it accepts "Path to the AGENTS.md file or a directory containing AGENTS.md"
(Location: src/context/mod.rs:123)

- Minor Issue 2: Improve error message for non-AGENTS.md file paths by
detecting file extensions early and returning explicit error message
"read_agents_md can only read AGENTS.md files" instead of degraded
"failed to read /some/path/README.md/AGENTS.md" error
(Location: src/context/mod.rs:139-150)

- Update test_read_agents_md_tool_invalid_filename to expect the new
improved error message

All context module tests (17 tests) pass successfully.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Update ReadAgentsMdTool::execute to handle both directory paths and full AGENTS.md
file paths. If path ends with AGENTS.md, read directly (backward compat). Otherwise,
treat as directory and append /AGENTS.md. Updated tool description to document both
path types.

Added tests for:
- test_read_agents_md_tool_with_directory_path (AC2.1)
- test_read_agents_md_tool_with_full_file_path (AC2.2)

Verifies v2-phase5.AC2.1 and v2-phase5.AC2.2

Use Path::strip_prefix(project_root) to convert absolute paths to relative paths
in resolve_agents_md output. Falls back to absolute path if stripping fails.
Updated tests verify paths are relative (e.g., 'src/auth/AGENTS.md').

Verifies v2-phase5.AC1.2 and v2-phase5.AC1.3

Replace extract_headings with extract_heading_summaries that returns Vec<(String, usize)>
with line counts for each top-level heading. Update resolve_agents_md to format summaries
as 'Heading (N lines), Heading2 (M lines)' format.

Verifies v2-phase5.AC1.1

Sidebar now polls projects every 10s so CLI-added projects appear in
the UI. Dashboard reacts to projectsState.projects without calling
loadProjects() itself, breaking the infinite re-render cycle that
caused flickering between spinner and loaded state.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Wrap router $effect in initRouter() called from App.svelte component
context to fix effect_orphan. Add resolve.conditions=['browser'] to
vite.config.ts so Vite picks the client entry point instead of the
server one (Svelte 5.50 changed package.json exports). Also add
justfile dev recipe for running daemon + web dev server together.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Add web/ domain AGENTS.md documenting the Svelte 5 SPA contracts,
dependencies, invariants, and routing. Update root AGENTS.md with
web UI in project structure, commands, and dependencies sections.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Updates .gitignore to exclude .vite/ and *.tsbuildinfo build artifacts.

Verified:
- bun run build produces clean output (only expected chunk size warning)
- npx tsc --noEmit passes with zero type errors
- bun run dev starts without errors
- All 8 views render correctly
- Placeholder component remains only for 404 fallback (correct)
- No unused imports found
- dist/ contains index.html and hashed JS/CSS bundles

- C1: Add FeedEvent type with feedId counter for stable Svelte keys in AgentMonitor
- C1: Change addEvent from unshift/pop to push/shift for newest-at-bottom behavior
- I1: Fix event feed ordering (completed with C1 push/shift change)
- I2: Make relativeTimeCounter reactive by passing to getEventTime function
- M1: Remove unused setQuery and clearSearch imports from GraphSearch
- M2: Remove empty unused $effect from GraphSearch
- M3: Remove dead 'created_at' check, use arrivedAt from FeedEvent

All tests pass. Build succeeds. Types check.

Implements Graph Search interface with full-text search capabilities:

SearchResult.svelte:
- Compact card showing node type badge, status badge, title, description
- Node type-specific colors (goal=purple, task=blue, decision=pink, etc.)
- Truncates description to 150 chars with ellipsis
- Click handler navigates to appropriate view based on node type:
- goal/task → task-tree view
- decision/option → decision-graph view
- observation/outcome/revisit → task-tree view
- Uses button element for a11y compliance

GraphSearch.svelte:
- Search bar with text input and search button (Enter key support)
- Filter bar with toggles for node types: All, Goals, Tasks, Decisions, Options,
Outcomes, Observations, Revisits
- Results list showing matching nodes with result count
- Empty states for no query, no results, or project not selected
- Requires project selection (prompts user to select first)
- Integrates with search store (query, results, nodeTypeFilter, loading, error)

Updates App.svelte to import and render GraphSearch for graph-search route

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Implements Session History view showing past sessions for a goal:
- Sessions list sorted by started_at descending (newest first)
- Session cards with start/end timestamps and duration
- Participating agents displayed as chips
- Expandable accordion to show full details
- Handoff notes rendered in preformatted text with monospace font
- Empty state message when no sessions found

Uses button element for session header with proper ARIA attributes
for a11y compliance (aria-expanded, role-implicit from button)

Updates App.svelte to import and render SessionHistory for
session-history route

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Implements Agent Monitor view showing real-time feed of WebSocket events:
- Active agents bar at top with status badges
- Event feed showing agent_spawned, agent_progress, agent_completed, tool_execution
- Auto-scroll to newest events with pause/resume toggle
- Clear feed button
- Relative time formatting (2s ago, 1m ago) updating every 10s
- Event type-specific styling and formatting
- Max 200 event items in feed ring buffer

Adds formatRelativeTime utility to date-formatting.ts

Updates App.svelte to import and render AgentMonitor for agent-monitor route

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Implements status badge component showing agent ID (truncated) and status with
colored backgrounds and pulsing/animated indicators for active states.

Status colors:
- spawning/initializing → pulsing blue
- working → animated green
- reporting → yellow
- completed → solid green
- failed → solid red

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Critical: Fix TypeScript type errors in decision-graph.test.ts by changing decisionStylesheet() return type to Array<cytoscape.StylesheetStyle> and adding proper type guards.

Important: Remove unused nowModeStatuses variable. Change getDisplayData() return type to Array<ElementDefinition>.

Minor: Add text-background-opacity to edge labels. Prevent double-render in CytoscapeGraph by adding initialized flag.

All verified: TypeScript check passes, 66 tests pass, build succeeds.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Tests the pure functions from decision-graph.ts:
- decisionsToElements: converts GraphNode/GraphEdge to Cytoscape elements
- filterNowMode: filters for active decisions and chosen options only
- decisionStylesheet: generates Cytoscape styling rules

Coverage:
- P4f.AC2.1: Decision nodes render as diamonds with gold color
- P4f.AC2.2: Option nodes render as hexagons (chosen=green, rejected=gray)
- P4f.AC2.3: Outcome and revisit nodes with distinct shapes and colors
- P4f.AC2.4: Edge labels show relationship types
- P4f.AC3.1: Now mode includes only active/decided decisions, chosen options
- P4f.AC3.2: History mode includes full evolution with rejected/abandoned nodes
- Edge case handling: empty input arrays, endpoint filtering, realistic scenarios

27 tests total, all passing. Verifies complete decision graph transformation pipeline.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Implements the Decision Graph view component that displays an interactive
graph of decision nodes (diamonds), option nodes (hexagons), outcomes
(ellipses), and revisit nodes (triangles) using Cytoscape.js with dagre layout.

Features:
- Now/History mode toggle for active-only vs full-evolution views
- Mode-based data loading via graph store (decisions vs decisionHistory)
- Detail panel showing selected node information with pros/cons for options
- Fit-to-view button for graph navigation
- Node count display
- Proper TypeScript typing using $derived for reactive updates

Integrates with:
- CytoscapeGraph wrapper component
- decision-graph transformation utilities
- GraphNodeCard for node detail display
- Graph store for data management
- Projects store for project selection

Updated App.svelte to route decision-graph requests to DecisionGraph component.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Implement decision-graph.ts with:
- decisionsToElements(): Convert GraphNode/GraphEdge to Cytoscape ElementDefinition[]
* Each node includes: id, label (title), type, status, description, priority, assigned_to, metadata
* Each edge includes: id, source, target, label, edgeType

- filterNowMode(): Filter for 'Now' mode (active decisions only)
* Include Decision nodes with status 'active' or 'decided'
* Include Option nodes with status 'chosen'
* Include Outcome nodes with status 'active' or 'completed'
* Exclude: rejected, abandoned, superseded nodes
* Prune edges where either endpoint is filtered out

- decisionStylesheet(): Return Cytoscape stylesheet with node-type styling
* Decision: diamond shape, gold (#FFD700) background, dark border
* Option chosen: hexagon, green (#32CD32)
* Option rejected: hexagon, gray (#666), dashed border
* Option abandoned: hexagon, muted red (#CD5C5C), dashed border
* Outcome completed: ellipse, green
* Outcome active: ellipse, blue (#4169E1)
* Revisit: triangle, orange (#FFA500)
* Edges: straight lines with arrow, chosen=thicker+green, rejected=dashed+gray
* Selected nodes: red border

TypeScript verification: passed

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Implement Svelte 5 wrapper around Cytoscape.js with:
- Props for elements, stylesheet, layout options, node/background click handlers
- Reactive updates when elements or stylesheet change
- Layout re-execution after elements update
- Export methods: fitView() and getInstance() for external control
- Performance optimizations: pixelRatio=1, wheelSensitivity=0.2
- Proper cleanup on unmount

Install cytoscape dependencies:
- cytoscape@3.33.1 (runtime)
- @types/cytoscape@3.31.0, cytoscape-dagre@2.5.0 (dev)

TypeScript verification: passed

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

- C1: Replace boolean isSelected prop with selectedNodeId to fix selection propagation through tree hierarchy. Each node now computes isSelected as selectedNodeId === node.id, preventing broken selection logic where child nodes were always deselected.
- C2: Fix NodeWithEdges type assumptions. Derive dependencies from outgoing_edges (filter by 'dependson' edge_type). Destructure [GraphEdge, GraphNode] tuples in #each loops for incoming/outgoing edges to access node titles and IDs.
- I1: Replace selectNode('') with clearSelection() on close button to avoid 404 errors from empty node ID API calls.
- I2: Change dependency indicator label from "blocked by" to "depends on" for semantic accuracy.
- M1: Replace !true with false for clarity in toggle initialization logic.

All fixes verified: TypeScript build ✓, tests pass (39/39) ✓, no type errors.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Implements Task 4 of Phase 5 (Task Tree View):
- New TaskTree.svelte view: three-panel layout with tree, detail, ready tasks
- New TreeNodeRow.svelte component: recursive tree rendering with expand/collapse
- TaskTree displays goal hierarchy from Contains edges, task statistics, progress bar
- Detail panel shows selected node with dependencies and edges
- Ready tasks panel at bottom with Next recommendation highlighting
- TreeNodeRow provides status/priority badges, agent assignments, dependency indicators
- All components use Svelte 5 runes (props, state, effect, derived)
- Fixed a11y issues in GraphNodeCard (role, tabindex, keyboard events)
- Integrated TaskTree into App.svelte route handler

Acceptance criteria met:
- P4e.AC1: Tree renders hierarchical goal/task/subtask structure with expandable nodes
- P4e.AC2: Status badges use color coding, dependency edges shown, ready tasks highlighted
- P4e.AC3: Ready tasks panel, next task recommendation, detail panel with full node info

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

- Tests buildTree hierarchical nesting, dependency identification, and ID sorting
- Tests flattenTree depth-first traversal
- Tests countTaskStats recursive status counting
- All 11 tests passing, including edge cases

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

- C1: Fix async effects in Dashboard/ProjectList/ProjectDetail
Replace $effect(async () => {...}) with sync $effect reading reactive deps,
then call separate async function (Svelte 5 doesn't support async in $effect)

- I1: Extract duplicate formatDate to web/src/lib/date-formatting.ts
Import from ProjectList.svelte and ProjectDetail.svelte

- I2: Parallelize API calls in Dashboard using Promise.allSettled
Fetch goals for all projects in parallel, then fetch agents in parallel

- M1: Remove unused ActiveAgent type import from Dashboard.svelte

- M2: Fix handleDecisionClick in ProjectDetail to extract goal ID
Use decisionId.split('.')[0] to get goal ID from hierarchical node ID

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Implements Phase 4d Task 4. Creates ProjectDetail.svelte to display project header with name, path, and registration date. Shows goal cards with title, status, priority, and task completion percentage. Each goal has action links to navigate to Tasks, Decisions, Agents, and Sessions views. Displays active decisions list. Handles loading, error, and empty states. Updates App.svelte to use ProjectDetail for project-detail route instead of placeholder.

Implements Phase 4d Task 3. Creates ProjectList.svelte to display all registered projects in a table with clickable project names that navigate to ProjectDetail. Shows project path in monospace and registration date in human-readable format. Handles loading, error, and empty states. Updates App.svelte to use ProjectList for project-list route instead of placeholder.

Implemented Dashboard view showing:
- Projects summary card with project names and goal counts (clickable to ProjectDetail)
- Active goals card listing all active goals across projects with status/priority badges
- Agents card showing total count of active agents

Dashboard loads projects and goals from API on mount using stores.
Handles loading and error states with LoadingSpinner and ErrorMessage components.
Responsive grid layout with dark theme matching design system.

Updated App.svelte to render Dashboard instead of placeholder for dashboard route.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Implemented four reusable UI components following Svelte 5 conventions:
- StatusBadge: colored badge for NodeStatus with color mapping
- PriorityBadge: colored indicator for Priority (renders nothing if null)
- LoadingSpinner: CSS-based spinner animation with configurable size
- ErrorMessage: styled error box with icon and message

All components use proper TypeScript types and follow house style conventions.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Critical (1):
- C1: loadGoals() discards API response — added goals field to graphState
and stored result: graphState.goals = await apiClient.listGoals(projectId)

Important (4):
- I1: App.svelte handleEvent callback now uses (event: WsEvent) type instead of (event: any)
and added proper import of WsEvent type
- I2: Sidebar.svelte isActive() no longer calls unused getCurrentRoute(), simplified to direct
comparison: return routerState.path === route
- I3: Created web/src/api/index.ts exporting shared apiClient singleton instance. Updated all 4
stores (projects, graph, agents, search) to import from '../api' instead of creating own
instances via createApiClient()
- I4: Sidebar.svelte $effect now uses loadAttempted flag to prevent infinite retry loop when
loadProjects() fails but projects list remains empty

Minor (3):
- M1: Placeholder.svelte changed interface Props to type Props for consistency
- M2: router.test.ts removed extra blank line between imports and describe block
- M3: App.svelte removed unused navigate import (already removed in I1 fix)

Verification:
- TypeScript: npx tsc --noEmit ✓
- Tests: bun run test → 28 passed ✓
- Build: bun run build → success ✓

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Implements Task 5 of Phase 3 (Stores + Router + App Shell):
- Replace App.svelte placeholder with full app shell component
- Add sidebar navigation with project selector and main nav links
- Implement view routing with Placeholder components for all routes
- Initialize WebSocket connection and dispatch events to graph/agents stores
- Use flexbox layout with fixed sidebar (240px) and scrollable main content
- Maintain dark theme with colors #12122a (sidebar) and #1a1a2e (main)

Also fixes Svelte 5 module-level $derived export issue:
- Convert selectedProject to getSelectedProject() function
- Convert goalNodes to getGoalNodes() function
- Convert tasksByStatus to getTasksByStatus() function

All files pass TypeScript check and build succeeds with no errors.
All router tests pass (14/14) and WebSocket tests pass (14/14).

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Implement a simple hash-based router using Svelte 5 runes with no external dependency:
- Pure exported matchRoute() function for unit testing route pattern matching
- routerState ($state) with path and params, automatically updated on hashchange events
- getCurrentRoute() function that matches current path against 8 route definitions
- navigate(path) function to update window.location.hash
- $effect setup/cleanup handles hashchange event listener registration and cleanup
- 14 comprehensive unit tests covering all routes and parameter extraction

The router supports these routes:
- / (dashboard)
- /projects (project-list)
- /projects/:projectId (project-detail)
- /goals/:goalId/tasks (task-tree)
- /goals/:goalId/decisions (decision-graph)
- /goals/:goalId/agents (agent-monitor)
- /goals/:goalId/sessions (session-history)
- /search (graph-search)

Verification: 28/28 tests pass, build succeeds, TypeScript types clean

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>