diff --git a/.daemon/logs/signet-2026-02-23.log b/.daemon/logs/signet-2026-02-23.log
index d4408289d..1514fdfc1 100644
--- a/.daemon/logs/signet-2026-02-23.log
+++ b/.daemon/logs/signet-2026-02-23.log
@@ -11468,3 +11468,23 @@
 {"timestamp":"2026-02-23T19:46:09.565Z","level":"info","category":"hooks","message":"Session start hook","data":{"harness":"claude-code","project":"/home/nicholai/signet/signetai"}}
 {"timestamp":"2026-02-23T19:46:09.992Z","level":"info","category":"hooks","message":"Session start completed","data":{"harness":"claude-code","project":"/home/nicholai/signet/signetai","memoryCount":19,"injectChars":11679,"injectPreview":"[memory active | /remember | /recall]\n\n# Current Date & Time\nMonday, February 23, 2026 at 12:46 PM (America/Denver)\n\n\n## Agent Instructions\n\nYou are Mr Claude, a helpful and thoughtful AI assistant.\n\nBehavioral Guidelines\n---\n\n- Be concise and direct\n- Ask clarifying questions when needed\n- Remember user preferences across sessions\n- Avoid sycophancy - be honest even when it's uncomfortable\n- Express nuanced judgment rather than hedging\n\n<!-- SIGNET:START -->\nSignet Agent System\n===\n\nYour identity and memory are managed by Signet, a portable agent identity\nsystem. This lets you maintain consistent behavior across different AI\nplatforms (Claude Code, OpenCode, Cursor, etc.).\n\nKey files in `~/.agents/`:\n- `agent.yaml` - Configuration\n- `AGENTS.md` - Instructions (this file)\n- `SOUL.md` - Personality and tone\n- `IDENTITY.md` - Agent identity\n- `USER.md` - User profile\n- `MEMORY.md` - Working memory summary\n\nDashboard: http://localhost:3850\n\nMemory\n---\n\nYou have access to persistent memory via Signet:\n\n```bash\nsignet remember \"User prefers dark mode and vim keybindings\"\nsignet recall \"user preferences\"\n```\n\nMemory is automatically loaded at session start. Important context is\nsummarized in `~/.agents/MEMORY.md`.\n\nSecrets\n---\n\nAPI keys and tokens are stored securely in Signet:\n\n```bash\nsignet secret get OPENAI_API_KEY\nsignet secret list\n```\n<!-- SIGNET:END -->\n\nAbout Your User\n---\n\nAdd information about yourself here so your agent knows who you are.\n\n- Name: \n- Timezone:\n- Preferences:\n\nProjects\n---\n\nList your active projects here.\n\n-\n\n## Soul\n\nsoul - persona & boundaries\n=========================\n\ntone and style\n---------\n\n- keep replies concise and direct\n- ask clarifying questions when needed\n- never send streaming/partial replies to external messaging surfaces\n\nformatting\n---------\n\nkeep markdown minimal. use ======== for main headings, ----- or ### if you\nreally need subheadings, but generally just stick to paragraphs.\n\n*italics* and **bold** are fine but use them sparingly - they're visually\nnoisy in neovim.\n\n- bullet points are okay\n- numbered lists are okay too\n\ncodeblocks ``` are fine, but get visually noisy when used too much.\n\nno excessive formatting. keep it clean and readable.\n\nreasoning\n---------\n\nfor every complex problem:\n1. decompose: break into sub-problems\n2. solve: address each problem with a confidence score (0.0-1.0)\n3. verify: check your logic, facts, completeness, and bias\n4. distill: combine using weighted confidence\n5. reflect: if conf\n...[truncated 9179 chars]","durationMs":427}}
 {"timestamp":"2026-02-23T19:46:09.993Z","level":"info","category":"watcher","message":"File changed","data":{"path":"/home/nicholai/.agents/memory/memories.db-wal"}}
+{"timestamp":"2026-02-23T19:46:15.054Z","level":"info","category":"git","message":"Auto-committed","data":{"message":"2026-02-23T19-46-14_auto_memory/memories.db-wal, memory/memories.db-wal, me","filesChanged":4}}
+{"timestamp":"2026-02-23T19:46:38.591Z","level":"info","category":"summary-worker","message":"Wrote session summary","data":{"path":"/home/nicholai/.agents/memory/2026-02-23-scheduled-agent-tasks-feature-plan.md","sessionKey":"9fbc5ee8-9703-4af2-aa57-bead58f15527","project":"/home/nicholai/signet/signetai","summaryChars":3133,"summaryPreview":"# 2026-02-23 Session Notes\n\n## Scheduled Agent Tasks Feature Plan\n\nImplementing a new \"Tasks\" dashboard tab that lets users schedule recurring agent prompts via cron expressions. The daemon will poll for due tasks and spawn CLI processes (Claude Code or OpenCode) to execute them.\n\n## Scope & Architecture\n\nFeature consists of seven integrated components:\n\n1. **Database**: Two new tables (`scheduled_tasks`, `task_runs`) with indices on `(enabled, next_run_at)` and `(task_id, status)`. Migration 012 with backwards-compat via index.ts registration.\n\n2. **Cron Parsing**: Thin wrapper module (`packages/daemon/src/scheduler/cron.ts`) around `cron-parser` dependency, exposing `computeNextRun()` and `validateCron()`. Dashboard provides preset shortcuts (15-min, hourly, daily 9am, weekly Monday, custom).\n\n3. **Scheduler Worker**: Polling worker at 15-second interval with max 3 concurrent spawned processes. Uses `Bun.spawn` for `claude --dangerously-skip-permissions` (Claude Code) and `opencode -m` (OpenCode) with 10-minute timeout per task. Transactional lease pattern: insert `task_runs` row and update `next_run_at` atomically before spawning. Startup cleanup marks stale running tasks as failed.\n\n4. **API Routes**: CRUD endpoints under `/api/tasks/*` — GET all, POST create, GET/PATCH/:id, DELETE, POST/:id/run (manual trigger), GET/:id/runs (paginated history).\n\n5. **Dashboard UI**: Kanban board with four columns (Scheduled, Running, Completed, Failed). Uses existing shadcn-svelte components (Table, Badge, Card, Sheet, Select, Input, Button, ScrollArea, Sonner). New components: TasksTab (header + board), TaskBoard, TaskCard, TaskForm (Sheet with Input/Textarea/Select), TaskDetail, RunLog. Requires adding Switch and Textarea shadcn components.\n\n6. **Integration**: Wires scheduler worker into daemon startup (line ~7002). Updates navigation store, app-sidebar, and routes in dashboard.\n\n7. **Documentation**: New SCHEDULING.md covering cron syntax, execution model, security, and troubleshooting. Updates to DASHBOARD.md, API.md, CLAUDE.md, and web/public/skill.md.\n\n## Key Decisions\n\n- **Polling granularity**: 15 seconds (cron minimum is 1 minute, no need for sub-minute precision)\n- **Concurrency limit**: 3 concurrent processes to prevent resource exhaustion\n- **Process spawning**: Via `Bun.spawn` with piped stdout/stderr, capped at 1MB per run\n- **Timeout**: 10-minute default per task execution\n- **Transactional safety**: Lease pattern prevents double-execution on scheduler restart\n- **Kanban columns**: Show task states (Scheduled/Running) and recent run states (Completed/Failed) — no drag-and-drop needed\n\n## Implementation Order\n\n1. Migration + type definitions\n2. Cron utilities\n3. Scheduler worker (spawn, worker, index modules)\n4. API routes wired into daemon.ts\n5. Dashboard: store, api client, navigation, components (sequential for context retention)\n6. Documentation\n\n## Open Threads\n\n- Confirm which shadcn-svelte components are already installed vs. need `npx shadcn-svelte@next add`\n- Test process timeout behavior and output truncation at 1MB\n- Define pagination limit for task_runs queries"}}
+{"timestamp":"2026-02-23T19:46:38.601Z","level":"info","category":"summary-worker","message":"Inserted session facts","data":{"total":12,"saved":12,"deduplicated":0,"factsPreview":["The scheduled_tasks table pre-computes next_run_at on creation and after each run — the scheduler worker queries WHERE enabled=1 AND next_run_at<=datetime('now') on each 15-second poll cycle","The scheduler worker uses a transactional lease pattern: inserts task_runs row and updates next_run_at atomically in a single withWriteTx before spawning the process, preventing double-execution on daemon restart","Process spawning for scheduled tasks uses Bun.spawn with 10-minute timeout, piped stdout/stderr capped at 1MB, and claudeCode invokes 'claude --dangerously-skip-permissions -p \"<prompt>\"' while opencode invokes 'opencode -m \"<prompt>\"'","Scheduler worker integrates into daemon.ts startup at line ~7002 (after startGitSyncTimer and initUpdateSystem) by calling startSchedulerWorker(getDbAccessor())","Dashboard Tasks tab uses a four-column kanban board: Scheduled (enabled tasks sorted by next_run_at), Running (live execution with elapsed timer), Completed (recent successful runs), Failed (recent failed runs). Columns 3-4 show runs, not t\n...[truncated 5 chars]","TasksTab dashboard components follow SkillsTab pattern with shadcn-svelte: existing Table, Badge, Card, Sheet, Select, Input, Button, ScrollArea, Sonner. Must add Switch and Textarea via 'npx shadcn-svelte@next add'","Cron preset table for dashboard: Every 15 min (*/15 * * * *), Hourly (0 * * * *), Daily 9am (0 9 * * *), Weekly Mon 9am (0 9 * * 1), plus custom user input validated with cron-parser","Scheduler worker enforces max 3 concurrent spawned processes and skips tasks that already have a 'running' run to prevent double-execution","Daemon startup cleanup runs UPDATE task_runs SET status='failed', error='daemon_restart' WHERE status='running' to mark stale running tasks from previous sessions","scheduled_tasks table schema: id, name, prompt, cron_expression, harness (claude-code|opencode), working_directory, enabled, last_run_at, next_run_at, created_at, updated_at with index on (enabled, next_run_at)"]}}
+{"timestamp":"2026-02-23T19:46:38.602Z","level":"info","category":"watcher","message":"File changed","data":{"path":"/home/nicholai/.agents/memory/memories.db-wal"}}
+{"timestamp":"2026-02-23T19:46:38.602Z","level":"info","category":"watcher","message":"File added","data":{"path":"/home/nicholai/.agents/memory/2026-02-23-scheduled-agent-tasks-feature-plan.md"}}
+{"timestamp":"2026-02-23T19:46:40.257Z","level":"info","category":"memory","message":"Memory saved","data":{"id":"01496365-f820-4afd-aa25-3ff0de746f83","type":"fact","pinned":false,"embedded":true}}
+{"timestamp":"2026-02-23T19:46:40.258Z","level":"info","category":"watcher","message":"File changed","data":{"path":"/home/nicholai/.agents/memory/memories.db-wal"}}
+{"timestamp":"2026-02-23T19:46:40.277Z","level":"info","category":"memory","message":"Memory saved","data":{"id":"da93cc4a-9d48-4d7e-a844-39355c980376","type":"fact","pinned":false,"embedded":true}}
+{"timestamp":"2026-02-23T19:46:40.293Z","level":"info","category":"memory","message":"Memory saved","data":{"id":"b4b9085a-8574-481c-8465-1d32d2bae4f7","type":"fact","pinned":false,"embedded":true}}
+{"timestamp":"2026-02-23T19:46:40.308Z","level":"info","category":"memory","message":"Memory saved","data":{"id":"8345aae8-831f-4054-8fcc-ae0051bf7dd0","type":"fact","pinned":false,"embedded":true}}
+{"timestamp":"2026-02-23T19:46:40.322Z","level":"info","category":"memory","message":"Memory saved","data":{"id":"96d5e36e-fe37-4c8b-943c-4494769cb8b3","type":"fact","pinned":false,"embedded":true}}
+{"timestamp":"2026-02-23T19:46:40.323Z","level":"info","category":"watcher","message":"Ingested memory file","data":{"path":"/home/nicholai/.agents/memory/2026-02-23-scheduled-agent-tasks-feature-plan.md","chunks":5,"sections":5,"filename":"2026-02-23-scheduled-agent-tasks-feature-plan"}}
+{"timestamp":"2026-02-23T19:46:40.323Z","level":"info","category":"watcher","message":"File changed","data":{"path":"/home/nicholai/.agents/memory/memories.db-wal"}}
+{"timestamp":"2026-02-23T19:46:40.900Z","level":"info","category":"watcher","message":"File changed","data":{"path":"/home/nicholai/.agents/memory/memories.db-wal"}}
+{"timestamp":"2026-02-23T19:46:44.511Z","level":"info","category":"hooks","message":"Session start hook","data":{"harness":"claude-code","project":"/home/nicholai/signet/signetai"}}
+{"timestamp":"2026-02-23T19:46:45.238Z","level":"info","category":"hooks","message":"Session start completed","data":{"harness":"claude-code","project":"/home/nicholai/signet/signetai","memoryCount":18,"injectChars":11701,"injectPreview":"[memory active | /remember | /recall]\n\n# Current Date & Time\nMonday, February 23, 2026 at 12:46 PM (America/Denver)\n\n\n## Agent Instructions\n\nYou are Mr Claude, a helpful and thoughtful AI assistant.\n\nBehavioral Guidelines\n---\n\n- Be concise and direct\n- Ask clarifying questions when needed\n- Remember user preferences across sessions\n- Avoid sycophancy - be honest even when it's uncomfortable\n- Express nuanced judgment rather than hedging\n\n<!-- SIGNET:START -->\nSignet Agent System\n===\n\nYour identity and memory are managed by Signet, a portable agent identity\nsystem. This lets you maintain consistent behavior across different AI\nplatforms (Claude Code, OpenCode, Cursor, etc.).\n\nKey files in `~/.agents/`:\n- `agent.yaml` - Configuration\n- `AGENTS.md` - Instructions (this file)\n- `SOUL.md` - Personality and tone\n- `IDENTITY.md` - Agent identity\n- `USER.md` - User profile\n- `MEMORY.md` - Working memory summary\n\nDashboard: http://localhost:3850\n\nMemory\n---\n\nYou have access to persistent memory via Signet:\n\n```bash\nsignet remember \"User prefers dark mode and vim keybindings\"\nsignet recall \"user preferences\"\n```\n\nMemory is automatically loaded at session start. Important context is\nsummarized in `~/.agents/MEMORY.md`.\n\nSecrets\n---\n\nAPI keys and tokens are stored securely in Signet:\n\n```bash\nsignet secret get OPENAI_API_KEY\nsignet secret list\n```\n<!-- SIGNET:END -->\n\nAbout Your User\n---\n\nAdd information about yourself here so your agent knows who you are.\n\n- Name: \n- Timezone:\n- Preferences:\n\nProjects\n---\n\nList your active projects here.\n\n-\n\n## Soul\n\nsoul - persona & boundaries\n=========================\n\ntone and style\n---------\n\n- keep replies concise and direct\n- ask clarifying questions when needed\n- never send streaming/partial replies to external messaging surfaces\n\nformatting\n---------\n\nkeep markdown minimal. use ======== for main headings, ----- or ### if you\nreally need subheadings, but generally just stick to paragraphs.\n\n*italics* and **bold** are fine but use them sparingly - they're visually\nnoisy in neovim.\n\n- bullet points are okay\n- numbered lists are okay too\n\ncodeblocks ``` are fine, but get visually noisy when used too much.\n\nno excessive formatting. keep it clean and readable.\n\nreasoning\n---------\n\nfor every complex problem:\n1. decompose: break into sub-problems\n2. solve: address each problem with a confidence score (0.0-1.0)\n3. verify: check your logic, facts, completeness, and bias\n4. distill: combine using weighted confidence\n5. reflect: if conf\n...[truncated 9201 chars]","durationMs":727}}
+{"timestamp":"2026-02-23T19:46:45.239Z","level":"info","category":"watcher","message":"File changed","data":{"path":"/home/nicholai/.agents/memory/memories.db-wal"}}
+{"timestamp":"2026-02-23T19:46:47.000Z","level":"info","category":"hooks","message":"Session start hook","data":{"harness":"claude-code","project":"/home/nicholai/signet/signetai"}}
+{"timestamp":"2026-02-23T19:46:47.376Z","level":"info","category":"hooks","message":"Session start completed","data":{"harness":"claude-code","project":"/home/nicholai/signet/signetai","memoryCount":18,"injectChars":11701,"injectPreview":"[memory active | /remember | /recall]\n\n# Current Date & Time\nMonday, February 23, 2026 at 12:46 PM (America/Denver)\n\n\n## Agent Instructions\n\nYou are Mr Claude, a helpful and thoughtful AI assistant.\n\nBehavioral Guidelines\n---\n\n- Be concise and direct\n- Ask clarifying questions when needed\n- Remember user preferences across sessions\n- Avoid sycophancy - be honest even when it's uncomfortable\n- Express nuanced judgment rather than hedging\n\n<!-- SIGNET:START -->\nSignet Agent System\n===\n\nYour identity and memory are managed by Signet, a portable agent identity\nsystem. This lets you maintain consistent behavior across different AI\nplatforms (Claude Code, OpenCode, Cursor, etc.).\n\nKey files in `~/.agents/`:\n- `agent.yaml` - Configuration\n- `AGENTS.md` - Instructions (this file)\n- `SOUL.md` - Personality and tone\n- `IDENTITY.md` - Agent identity\n- `USER.md` - User profile\n- `MEMORY.md` - Working memory summary\n\nDashboard: http://localhost:3850\n\nMemory\n---\n\nYou have access to persistent memory via Signet:\n\n```bash\nsignet remember \"User prefers dark mode and vim keybindings\"\nsignet recall \"user preferences\"\n```\n\nMemory is automatically loaded at session start. Important context is\nsummarized in `~/.agents/MEMORY.md`.\n\nSecrets\n---\n\nAPI keys and tokens are stored securely in Signet:\n\n```bash\nsignet secret get OPENAI_API_KEY\nsignet secret list\n```\n<!-- SIGNET:END -->\n\nAbout Your User\n---\n\nAdd information about yourself here so your agent knows who you are.\n\n- Name: \n- Timezone:\n- Preferences:\n\nProjects\n---\n\nList your active projects here.\n\n-\n\n## Soul\n\nsoul - persona & boundaries\n=========================\n\ntone and style\n---------\n\n- keep replies concise and direct\n- ask clarifying questions when needed\n- never send streaming/partial replies to external messaging surfaces\n\nformatting\n---------\n\nkeep markdown minimal. use ======== for main headings, ----- or ### if you\nreally need subheadings, but generally just stick to paragraphs.\n\n*italics* and **bold** are fine but use them sparingly - they're visually\nnoisy in neovim.\n\n- bullet points are okay\n- numbered lists are okay too\n\ncodeblocks ``` are fine, but get visually noisy when used too much.\n\nno excessive formatting. keep it clean and readable.\n\nreasoning\n---------\n\nfor every complex problem:\n1. decompose: break into sub-problems\n2. solve: address each problem with a confidence score (0.0-1.0)\n3. verify: check your logic, facts, completeness, and bias\n4. distill: combine using weighted confidence\n5. reflect: if conf\n...[truncated 9201 chars]","durationMs":376}}
+{"timestamp":"2026-02-23T19:46:47.376Z","level":"info","category":"watcher","message":"File changed","data":{"path":"/home/nicholai/.agents/memory/memories.db-wal"}}
diff --git a/memory/2026-02-23-scheduled-agent-tasks-feature-plan.md b/memory/2026-02-23-scheduled-agent-tasks-feature-plan.md
new file mode 100644
index 000000000..77eaa4d97
--- /dev/null
+++ b/memory/2026-02-23-scheduled-agent-tasks-feature-plan.md
@@ -0,0 +1,47 @@
+# 2026-02-23 Session Notes
+
+## Scheduled Agent Tasks Feature Plan
+
+Implementing a new "Tasks" dashboard tab that lets users schedule recurring agent prompts via cron expressions. The daemon will poll for due tasks and spawn CLI processes (Claude Code or OpenCode) to execute them.
+
+## Scope & Architecture
+
+Feature consists of seven integrated components:
+
+1. **Database**: Two new tables (`scheduled_tasks`, `task_runs`) with indices on `(enabled, next_run_at)` and `(task_id, status)`. Migration 012 with backwards-compat via index.ts registration.
+
+2. **Cron Parsing**: Thin wrapper module (`packages/daemon/src/scheduler/cron.ts`) around `cron-parser` dependency, exposing `computeNextRun()` and `validateCron()`. Dashboard provides preset shortcuts (15-min, hourly, daily 9am, weekly Monday, custom).
+
+3. **Scheduler Worker**: Polling worker at 15-second interval with max 3 concurrent spawned processes. Uses `Bun.spawn` for `claude --dangerously-skip-permissions` (Claude Code) and `opencode -m` (OpenCode) with 10-minute timeout per task. Transactional lease pattern: insert `task_runs` row and update `next_run_at` atomically before spawning. Startup cleanup marks stale running tasks as failed.
+
+4. **API Routes**: CRUD endpoints under `/api/tasks/*` — GET all, POST create, GET/PATCH/:id, DELETE, POST/:id/run (manual trigger), GET/:id/runs (paginated history).
+
+5. **Dashboard UI**: Kanban board with four columns (Scheduled, Running, Completed, Failed). Uses existing shadcn-svelte components (Table, Badge, Card, Sheet, Select, Input, Button, ScrollArea, Sonner). New components: TasksTab (header + board), TaskBoard, TaskCard, TaskForm (Sheet with Input/Textarea/Select), TaskDetail, RunLog. Requires adding Switch and Textarea shadcn components.
+
+6. **Integration**: Wires scheduler worker into daemon startup (line ~7002). Updates navigation store, app-sidebar, and routes in dashboard.
+
+7. **Documentation**: New SCHEDULING.md covering cron syntax, execution model, security, and troubleshooting. Updates to DASHBOARD.md, API.md, CLAUDE.md, and web/public/skill.md.
+
+## Key Decisions
+
+- **Polling granularity**: 15 seconds (cron minimum is 1 minute, no need for sub-minute precision)
+- **Concurrency limit**: 3 concurrent processes to prevent resource exhaustion
+- **Process spawning**: Via `Bun.spawn` with piped stdout/stderr, capped at 1MB per run
+- **Timeout**: 10-minute default per task execution
+- **Transactional safety**: Lease pattern prevents double-execution on scheduler restart
+- **Kanban columns**: Show task states (Scheduled/Running) and recent run states (Completed/Failed) — no drag-and-drop needed
+
+## Implementation Order
+
+1. Migration + type definitions
+2. Cron utilities
+3. Scheduler worker (spawn, worker, index modules)
+4. API routes wired into daemon.ts
+5. Dashboard: store, api client, navigation, components (sequential for context retention)
+6. Documentation
+
+## Open Threads
+
+- Confirm which shadcn-svelte components are already installed vs. need `npx shadcn-svelte@next add`
+- Test process timeout behavior and output truncation at 1MB
+- Define pagination limit for task_runs queries
\ No newline at end of file
diff --git a/memory/memories.db-shm b/memory/memories.db-shm
index 245e7c18a..053fb15a4 100644
Binary files a/memory/memories.db-shm and b/memory/memories.db-shm differ
diff --git a/memory/memories.db-wal b/memory/memories.db-wal
index afc38e3cc..66b2e6246 100644
Binary files a/memory/memories.db-wal and b/memory/memories.db-wal differ