0.5.0

9to5

Automated agents for Claude Code.

Agents that run on your schedule, from a webhook, or on demand. They stay under budget and report back to your inbox.

Agents: for complete option references and scheduling examples, read skills/9to5/SKILL.md or install the skill with npx skills add Michaelliv/9to5. You should run 9to5 onboard to add instructions to ~/.claude/CLAUDE.md.

What people automate

The examples/ directory has ready-to-import agents:

9to5 agent import examples/morning-briefing.json

Install

Requires Claude Code installed and authenticated.

npm install -g 9to5

Or run without installing:

npx 9to5 <command>

Quick start

# Create an agent that runs daily at 9am
9to5 agent add "morning-review" \
  --prompt "Review yesterday's commits and summarize changes" \
  --rrule "FREQ=DAILY;BYHOUR=9" \
  --model sonnet \
  --max-budget-usd 0.25

# Run it now to see what you get
9to5 agent run <id>

# Browse everything in the TUI
9to5 ui

Why not just cron + claude -p?

You could. 9to5 adds what you'd end up building yourself:

• Budget caps per agent - it won't burn your API credits overnight
• Run history with cost and duration - see what each agent cost and how long it worked
• Inbox - read/unread notifications so you know what happened while you were away
• Session resume - pick up where an agent left off with resume <run-id>
• Interactive TUI - browse, run, pause, delete, and drill into output without leaving the terminal. The daemon auto-starts and self-heals.
• Export/import - share agents as JSON, bring them to another machine
• Model and system prompt per agent - different personas for different jobs

Commands

Agent management (9to5 agent)

Runs & inbox

Daemon (9to5 daemon)

Webhooks (9to5 webhook)

Other

TUI dashboard

Launch with 9to5 ui for a two-panel terminal dashboard:

• Agents - browse, run, pause, and delete with a detail panel showing prompt, schedule, and config
• Runs - drill into an agent to see execution history with status, duration, cost, and structured output
• Hotkeys - r run, p pause/resume, dd delete (with u to undo), c copy output, ⏎ toggle read, q quit

Webhooks

Trigger agents from GitHub Actions, CI, Zapier, or any script. Two trigger paths — local HTTP and remote via ntfy.sh — both secured with HMAC-SHA256 signing. No infrastructure, no dependencies, completely free.

Webhooks are enabled by default — the daemon auto-generates a secret on first start. Use webhook disable to turn them off.

# View webhook config and URLs
9to5 webhook info

# Get ready-to-use curl commands for an agent
9to5 webhook url <agent-id>

# Regenerate the secret (breaks existing integrations)
9to5 webhook refresh

Local trigger — POST to http://localhost:9505/trigger/<id> with a signed body and X-Signature header. Works for scripts on the same machine or in your local network.

Remote trigger — POST to the ntfy.sh URL shown by webhook info. The daemon subscribes via SSE. Works from anywhere — CI, GitHub Actions, other machines.

Both paths verify HMAC-SHA256 signatures and reject messages older than 5 minutes.

# Example: trigger from a script
SECRET='<your-webhook-secret>'
BODY='{"automation_id":"<id>","ts":'$(date +%s000)'}'
SIG=$(echo -n "$BODY" | openssl dgst -sha256 -hmac "$SECRET" | sed 's/^.* //')

# Local
curl -X POST http://localhost:9505/trigger/<id> \
  -H "Content-Type: application/json" \
  -H "X-Signature: $SIG" \
  -d "$BODY"

# Remote (ntfy)
curl -X POST <ntfy-url> \
  -H "Content-Type: application/json" \
  -d '{"payload":"'"$BODY"'","sig":"'"$SIG"'"}'

Data

All data is stored locally in ~/.9to5/ (SQLite database, daemon PID file, and webhook secret).

Docs

Landing page at michaelliv.github.io/9to5. For agent-friendly option references and scheduling examples, see skills/9to5/SKILL.md.

License

MIT