Scheduled Tasks

Schedule recurring agent prompts that the Signet Daemon executes automatically via Claude Code or OpenCode CLI.

Overview

Scheduled tasks let you automate recurring agent workflows — PR reviews, code linting, status summaries, dependency checks, etc. The daemon evaluates cron expressions and spawns CLI processes on schedule.

Source code: packages/daemon/src/scheduler/

Creating Tasks

Via Dashboard

1. Open the Signet Dashboard (http://localhost:3850)
2. Navigate to the Tasks tab
3. Click + New Task
4. Fill in the form:
   • Name: descriptive label (e.g. “Review open PRs”)
   • Prompt: what the agent should do
   • Harness: Claude Code or OpenCode
   • Schedule: pick a preset or enter a custom cron expression
   • Working Directory: optional project path for context
5. Click Create Task

Via API

curl -X POST http://localhost:3850/api/tasks \
  -H "Content-Type: application/json" \
  -d '{
    "name": "Daily PR review",
    "prompt": "Review all open pull requests and summarize findings",
    "cronExpression": "0 9 * * *",
    "harness": "claude-code",
    "workingDirectory": "/home/user/my-project"
  }'

Required fields: name, prompt, cronExpression, harness.

Optional fields: workingDirectory.

Supported harness values: claude-code, opencode.

Cron Expressions

Standard 5-field cron syntax: minute hour day-of-month month day-of-week

Presets

The dashboard offers these built-in presets (defined in packages/daemon/src/scheduler/cron.ts):

Custom expressions are validated before saving using cron-parser. The validateCron() function returns true for valid expressions and false otherwise — invalid expressions are rejected at creation time.

Execution Model

• The daemon polls every 15 seconds for due tasks
• Maximum 3 concurrent task processes
• Each run gets a unique UUID and captures stdout/stderr
• Output is capped at 1 MB per stream (1,048,576 characters)
• Default timeout: 10 minutes per task
• Tasks that are already running are skipped (no double-execution)
• On daemon restart, any in-progress runs are marked as failed

Process Commands

The spawnTask() function in packages/daemon/src/scheduler/spawn.ts builds the CLI command based on the harness:

• Claude Code: claude --dangerously-skip-permissions -p "<prompt>"
• OpenCode: opencode run --format json "<prompt>"

Before spawning, the function checks that the CLI binary exists on PATH via Bun.which(). If the binary isn’t found, the run fails immediately with a "CLI binary not found on PATH" error.

Environment Isolation

Spawned processes inherit the daemon’s environment with two modifications:

• CLAUDECODE is stripped to avoid nested-session detection
• SIGNET_NO_HOOKS is set to "1" to prevent hook loops (the spawned agent shouldn’t trigger Signet hooks back into the daemon)

Timeout Behavior

When a task exceeds its timeout:

1. SIGTERM is sent to the process
2. After 5 additional seconds, SIGKILL is sent if still alive
3. The run is recorded with timedOut: true and an error message

Startup Recovery

When the daemon starts, the scheduler marks all pending and running task runs as failed with error "daemon_restart". This prevents orphaned runs from blocking future executions.

Task Streaming

The daemon provides real-time output streaming for running tasks via Server-Sent Events (SSE).

Endpoint

GET /api/tasks/:id/stream

Returns an SSE stream with the following event types:

Event Types

connected — Sent immediately on connection.

{
  "type": "connected",
  "taskId": "abc-123",
  "timestamp": "2026-03-01T10:00:00.000Z"
}

run-started — A new run has begun. Also sent as a replay if a run is already in progress when the client connects.

{
  "type": "run-started",
  "taskId": "abc-123",
  "runId": "def-456",
  "startedAt": "2026-03-01T10:00:01.000Z",
  "timestamp": "2026-03-01T10:00:01.000Z"
}

run-output — A chunk of stdout or stderr from the running process.

{
  "type": "run-output",
  "taskId": "abc-123",
  "runId": "def-456",
  "stream": "stdout",
  "chunk": "Analyzing pull requests...\n",
  "timestamp": "2026-03-01T10:00:05.000Z"
}

run-completed — The run has finished.

{
  "type": "run-completed",
  "taskId": "abc-123",
  "runId": "def-456",
  "status": "completed",
  "completedAt": "2026-03-01T10:05:00.000Z",
  "exitCode": 0,
  "error": null,
  "timestamp": "2026-03-01T10:05:00.000Z"
}

Replay on Connect

When a client connects while a task is already running, the stream replays the current run state: a run-started event followed by buffered stdout/stderr chunks (up to 200,000 characters per stream). This lets late-joining clients catch up without missing output.

Buffer Management

The in-memory buffer retains the most recent 200,000 characters per stream (stdout and stderr independently). When the buffer exceeds this limit, the oldest chunks are trimmed from the front. The buffer is cleared when a run completes.

API Endpoints

Managing Tasks

Enable/Disable

Toggle the switch on any task card in the dashboard, or via API:

curl -X PATCH http://localhost:3850/api/tasks/<id> \
  -H "Content-Type: application/json" \
  -d '{"enabled": false}'

Manual Run

Trigger a task immediately without waiting for the next scheduled time. Click “Run Now” in the task detail panel or:

curl -X POST http://localhost:3850/api/tasks/<id>/run

Viewing Run History

Click any task card in the dashboard to see its run history with stdout/stderr output. Or via API:

curl http://localhost:3850/api/tasks/<id>/runs?limit=20&offset=0

Error Handling and Retry

Tasks do not automatically retry on failure. Each run is a one-shot execution. If a run fails:

1. The run status is set to "failed"
2. The error field captures the reason (timeout, non-zero exit, spawn error)
3. The task’s next_run_at is still advanced to the next cron tick
4. The task remains enabled and will execute again at the next scheduled time

A run is considered failed when:

• The process exits with a non-zero exit code
• The error field on the SpawnResult is non-null (binary not found, spawn error)
• The process times out

Successful runs have status: "completed" and exitCode: 0.

Security

Claude Code runs with --dangerously-skip-permissions, meaning tasks execute without user approval gates. The dashboard displays a warning when creating Claude Code tasks.

Only schedule tasks you trust. The daemon runs them with the same permissions as the daemon process itself.

Troubleshooting

Task not running?

• Check that the daemon is running (signet status)
• Verify the CLI binary is on PATH (which claude or which opencode)
• Check the task is enabled in the dashboard

Task failing?

• Open the task detail to view stdout/stderr from the last run
• Check for timeout issues (default 10 minutes)
• Verify the working directory exists and is accessible

Daemon restart clears running tasks?

• This is expected — in-progress runs are marked as failed on restart
• The task will be picked up again at the next scheduled time