Daemon Guide

The background engine that keeps claude-ops fast.

Note

ops-daemon is a single launchd-managed process that supervises seven sub-services. It replaced per-service launchd agents in v0.5.0, then gained briefing-pre-warm (v0.6.0) and early-install (v0.8.0).

What the daemon does

flowchart LR
    D[ops-daemon] --> B[briefing-pre-warm<br/>every 2 min]
    D --> W[wacli-sync<br/>continuous]
    D --> M[memory-extractor<br/>every 30 min]
    D --> I[inbox-digest<br/>every 4 hours]
    D --> S[store-health<br/>daily 9am]
    D --> C[competitor-intel<br/>weekly Mon 10am]
    D --> L[message-listener<br/>continuous]

    B --> BC[daemon-cache.json]
    W --> WH[~/.wacli/.health]
    M --> ME[memories/]
    I --> IL[inbox log]
    S --> SR[store-health report]
    C --> CR[competitor report]
    L --> LL[message log]

    classDef svc fill:#6366f1,color:#fff,stroke:#4338ca
    classDef out fill:#22c55e,color:#fff,stroke:#15803d
    classDef root fill:#8b5cf6,color:#fff,stroke:#6d28d9
    class D root
    class B,W,M,I,S,C,L svc
    class BC,WH,ME,IL,SR,CR,LL out

Service	Cadence	Purpose
`briefing-pre-warm`	every 2 min	Runs `bin/ops-gather` and caches dashboard data for `/ops:go`
`wacli-sync`	continuous	Keeps WhatsApp Web connection alive, auto-backfills `@lid` chats
`memory-extractor`	every 30 min	Haiku 4.5 summarizes new wacli + gog data into `memories/`
`inbox-digest`	every 4 hours	Ranks unread messages across channels, writes digest
`store-health`	daily at 09:00	Shopify inventory / order / fulfillment sanity check
`competitor-intel`	weekly Mon 10:00	Scrapes competitor signals, writes weekly brief
`message-listener`	continuous	Polls wacli/Telegram for new messages, writes to local event log

Important

All services are local-only. message-listener never sends messages outbound on its own — it only writes to a log that the daemon reads. See Privacy and Security for the exact list of network targets.

Early install strategy (v0.8.0)

Pre-v0.8.0 the daemon was installed at Step 5b — after channels, MCPs, and registry. That meant users waited ~10s for their first /ops:go because the cache was empty.

v0.8.0 moved daemon install to Step 2c, immediately after CLIs. The briefing-pre-warm service starts working the moment the plist loads, and continues to warm the cache while the rest of the setup wizard runs.

sequenceDiagram
    autonumber
    participant U as User
    participant W as Setup Wizard
    participant D as ops-daemon
    participant G as ops-gather
    participant C as /ops:go

    U->>W: /ops:setup
    W->>W: Step 1 · Pick sections
    W->>W: Step 2 · Install CLIs
    W->>D: Step 2c · Install daemon (early!)
    activate D
    D->>G: briefing-pre-warm tick #1
    G-->>D: registry, PRs, CI, unread, revenue...
    Note over W,D: Wizard continues in parallel
    W->>W: Step 3 · Configure channels (60–180s)
    D->>G: briefing-pre-warm tick #2 (+2 min)
    W->>W: Step 4 · MCPs · 5 · Registry · 6 · Prefs
    D->>G: briefing-pre-warm tick #N
    W->>W: Step 5b · Reconcile daemon services
    W-->>U: Setup complete
    U->>C: /ops:go
    C->>D: read daemon-cache.json
    D-->>C: cached data (<3s)
    deactivate D

Tip

By the time the wizard finishes, the cache has 3–8 warm-up cycles banked. /ops:go loads from cache in under 3 seconds instead of cold-gathering in 10+.

Health file contract

Two health files, one per concern.

`~/.wacli/.health`

Written by wacli-sync on every successful poll.

{
  "status": "ok",
  "last_sync": "2026-04-14T07:45:00Z",
  "wacli_pid": 12345,
  "message_count": 1420
}

Field	Meaning
`status`	`ok` · `degraded` · `down`
`last_sync`	ISO timestamp of last successful WhatsApp poll
`wacli_pid`	PID of the running `wacli` process
`message_count`	Total messages in local SQLite DB

`daemon-health.json`

Written by the supervisor. Lives at ~/.claude/plugins/data/ops-ops-marketplace/daemon-health.json.

{
  "supervisor_pid": 54321,
  "uptime_seconds": 87442,
  "services": {
    "briefing-pre-warm": { "status": "ok", "last_run": "2026-04-14T07:44:12Z" },
    "wacli-sync":        { "status": "ok", "last_run": "2026-04-14T07:45:00Z" },
    "memory-extractor":  { "status": "ok", "last_run": "2026-04-14T07:30:04Z" },
    "inbox-digest":      { "status": "ok", "last_run": "2026-04-14T04:00:00Z" },
    "store-health":      { "status": "ok", "last_run": "2026-04-14T09:00:00Z" },
    "competitor-intel":  { "status": "idle", "next_run": "2026-04-21T10:00:00Z" },
    "message-listener":  { "status": "ok", "last_event": "2026-04-14T07:44:58Z" }
  }
}

PreToolUse hook — `ops-pretool-wacli-health`

hooks/whatsapp-health-check.sh (registered in hooks.json) fires before any wacli command. It reads ~/.wacli/.health and:

Condition	Behavior
`status === "ok"` AND `last_sync < 10 min`	Proceeds silently
`status === "degraded"` OR `last_sync >= 10 min`	Surfaces warning + restart instructions
Health file missing	Surfaces "daemon not running" + install instructions

Warning

The hook fails closed. Skills warn instead of silently retrying with stale auth. This is intentional — per Rule 3 and the privacy policy, users must see every health regression rather than get misleading "no new messages" results.

Brain layer · briefing cache

bin/ops-brain runs inside briefing-pre-warm. Every 2 minutes it:

Calls bin/ops-gather — which probes registry projects, GitHub PRs, CI, AWS ECS, Shopify, Klaviyo, Meta Ads, unread counts, GSD phases
Writes the result to daemon-cache.json
Detects urgent patterns (your name mentions, "urgent", "ASAP", "fire", "down", "broken") in unread messages
Flips the urgent_flag in daemon-health.json so /ops:go can surface them instantly

Result: /ops:go reads one cached JSON file instead of spawning 8 parallel gather scripts.

Service reconciliation (Step 5b)

After channels + registry are configured, Step 5b reconciles the daemon:

Adds wacli-sync if WhatsApp is configured
Adds message-listener if WhatsApp OR Telegram is configured
Adds inbox-digest if >=2 channels are configured
Adds store-health if Shopify is configured
Adds competitor-intel if /ops:marketing is selected
Verifies all services launch, prints a green-check summary

Because the daemon was installed at Step 2c with a minimal service list, Step 5b's job is to add channel-dependent services — not reinstall everything from scratch.

Start · stop · restart

launchd (recommended — survives reboots)

# Start (first-time install or after edits)
launchctl load ~/Library/LaunchAgents/com.claude-ops.daemon.plist

# Stop
launchctl unload ~/Library/LaunchAgents/com.claude-ops.daemon.plist

# Force restart (common fix for stuck services)
launchctl kickstart -k gui/$UID/com.claude-ops.daemon

# Check status
launchctl list | grep claude-ops

Foreground (debug mode)

~/.claude/plugins/cache/ops-marketplace/ops/<version>/scripts/ops-daemon.sh start
~/.claude/plugins/cache/.../scripts/ops-daemon.sh status
~/.claude/plugins/cache/.../scripts/ops-daemon.sh stop

Via the plugin

/ops:setup daemon     # re-install, regenerate plist with current paths
/ops:doctor           # auto-repair (runs ops-autofix + kicks the daemon)

Logs

Primary log location:

~/.claude/plugins/data/ops-ops-marketplace/logs/ops-daemon.log

Follow live:

tail -f ~/.claude/plugins/data/ops-ops-marketplace/logs/ops-daemon.log

Filter by service:

grep briefing-pre-warm ~/.claude/plugins/data/ops-ops-marketplace/logs/ops-daemon.log
grep memory-extractor  ~/.claude/plugins/data/ops-ops-marketplace/logs/ops-daemon.log

Note

Logs are rotated at 10 MB with 3 backup files kept. Nothing in the log file is ever transmitted off-machine — see Privacy and Security.

What the daemon does NOT do

Explicit list of non-behaviors

No telemetry. No phone-home, no crash reports, no usage metrics.
No outbound network except to the user's own configured APIs (Anthropic for Haiku calls, WhatsApp's servers via wacli, the user's Shopify store, etc.)
No sending messages unless a skill (running in an interactive Claude Code session) explicitly calls send_message with user confirmation.
No disk-wide scans. Every read is a targeted path: registry projects, ~/.wacli/, ~/.claude/plugins/data/ops-ops-marketplace/.
No credential harvesting post-setup. Auto-scan runs only inside /ops:setup.

Troubleshooting cross-reference

Symptom	See
`/ops:go` is slow (>10s)	Troubleshooting#daemon — pre-warm cache not updating
`launchctl load` fails	Troubleshooting#daemon — launchd errors
`wacli health degraded` warning	Troubleshooting#daemon — health file stale
Memories not extracting	Troubleshooting#memories-not-extracting + Memories System
Still stuck	Run `/ops:doctor --verbose`

Daemon Guide

Daemon Guide

What the daemon does

Early install strategy (v0.8.0)

Health file contract

~/.wacli/.health

daemon-health.json

PreToolUse hook — ops-pretool-wacli-health

Brain layer · briefing cache

Service reconciliation (Step 5b)

Start · stop · restart

launchd (recommended — survives reboots)

Foreground (debug mode)

Via the plugin

Logs

What the daemon does NOT do

Troubleshooting cross-reference

See also

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

claude-ops Wiki

Getting started

Reference

v2.0 — autonomy layer

v2.4 — smart home

v2.3 — competitor intel

v2.11 — telegram bot push · rotation improvements

v2.9 — voice & phone

Usage

Internals

Clone this wiki locally

`~/.wacli/.health`

`daemon-health.json`

PreToolUse hook — `ops-pretool-wacli-health`