Strix/.claude/skills/add_probe_detector_strix/SKILL.md

name, description, disable-model-invocation, argument-hint

name	description	disable-model-invocation	argument-hint
add_probe_detector_strix	Add a new device type detector to the Strix probe system. Covers adding new probers, result types, and detector functions.	true	detector-name

Add Probe Detector to Strix

You are adding a new device type detector to the Strix probe system. The probe system runs when a user enters an IP address -- it discovers what's at that IP and determines the device type. The device type drives the frontend flow.

The detector name is provided as argument (e.g. /add_probe_detector_strix onvif). If no argument, use AskUserQuestion to ask which detector to add.

Repository

• Strix: /home/user/Strix
• go2rtc (reference): /home/user/go2rtc

---

STEP 0: Understand the probe system

Before writing anything, read these files COMPLETELY:

/home/user/Strix/internal/probe/probe.go   -- glue: Init(), runProbe(), detectors, API handler
/home/user/Strix/pkg/probe/models.go       -- all data structures (Response, Probes, result types)
/home/user/Strix/pkg/probe/ping.go         -- prober example: ICMP ping
/home/user/Strix/pkg/probe/ports.go        -- prober example: TCP port scan
/home/user/Strix/pkg/probe/arp.go          -- prober example: ARP lookup
/home/user/Strix/pkg/probe/dns.go          -- prober example: reverse DNS
/home/user/Strix/pkg/probe/http.go         -- prober example: HTTP HEAD request
/home/user/Strix/pkg/probe/mdns.go         -- prober example: HomeKit mDNS query
/home/user/Strix/pkg/probe/oui.go          -- prober example: OUI vendor lookup

Read ALL of them. Every prober is different. Understand the full picture before proceeding.

How the probe system works

The probe has three layers:

Layer 1: Probers (pkg/probe/)

Pure functions that gather raw data about an IP address. Each runs in parallel with a shared 100ms timeout context. They do NOT interpret results -- just collect facts.

Current probers:

• Ping() -- ICMP echo, returns latency
• ScanPorts() -- TCP connect to all known camera ports, returns open ports
• ReverseDNS() -- reverse DNS lookup, returns hostname
• LookupARP() -- reads /proc/net/arp, returns MAC address
• LookupOUI() -- looks up MAC prefix in SQLite, returns vendor name
• ProbeHTTP() -- HTTP HEAD to ports 80/8080, returns status + server header
• QueryHAP() -- mDNS query for HomeKit Accessory Protocol, returns device info

Every prober writes its result into resp.Probes.{Name} via mutex.

Layer 2: Detectors (internal/probe/probe.go)

Functions registered in the detectors slice. They run AFTER all probers complete. Each detector receives the full *probe.Response with all probe results and returns a device type string (or empty string to pass).

var detectors []func(*probe.Response) string

Detectors are checked in order. First non-empty result wins and sets resp.Type.

Default type is "standard". If device is unreachable, type is "unreachable".

Layer 3: API (internal/probe/probe.go)

GET /api/probe?ip=192.168.1.100 returns the full Response JSON. The frontend uses type field to decide which UI flow to show.

Data flow

IP address
  |
  v
[All probers run in parallel, 100ms timeout]
  |
  v
probe.Response filled with results
  |
  v
[Detectors run in order on the Response]
  |
  v
resp.Type = "homekit" | "standard" | "unreachable" | ...
  |
  v
JSON response to frontend

API response example

{
  "ip": "192.168.1.100",
  "reachable": true,
  "latency_ms": 2.5,
  "type": "homekit",
  "probes": {
    "ping": {"latency_ms": 2.5},
    "ports": {"open": [80, 554, 5353]},
    "dns": {"hostname": "camera.local"},
    "arp": {"mac": "C0:56:E3:AA:BB:CC", "vendor": "Hikvision"},
    "mdns": {
      "name": "My Camera",
      "device_id": "AA:BB:CC:DD:EE:FF",
      "model": "Camera 1080p",
      "category": "camera",
      "paired": false,
      "port": 80
    },
    "http": {"port": 80, "status_code": 200, "server": "nginx"}
  }
}

---

STEP 1: Determine what you need

Use AskUserQuestion to discuss with the user. There are two scenarios:

Scenario A: Detector only (using existing probe data)

The detector can determine device type from data already collected by existing probers. No new prober needed.

Examples:

• Detect ONVIF cameras by checking if port 80 is open and HTTP server header contains "onvif" or specific vendor strings
• Detect specific brands by ARP vendor name
• Detect UPnP devices by checking specific open ports

In this case: skip to STEP 3.

Scenario B: New prober + detector

Need to collect new data that existing probers don't provide. Requires adding a new prober to pkg/probe/ and a new result type to models.go.

Examples:

• ONVIF discovery (send ONVIF GetCapabilities request)
• UPnP SSDP discovery
• Specific protocol handshake

In this case: proceed to STEP 2.

---

STEP 2: Add new prober (Scenario B only)

2a: Add result type to models.go

Edit /home/user/Strix/pkg/probe/models.go:

1. Add new result struct:

type {Name}Result struct {
    // fields specific to this probe
}

2. Add field to Probes struct:

type Probes struct {
    Ping  *PingResult  `json:"ping"`
    Ports *PortsResult `json:"ports"`
    DNS   *DNSResult   `json:"dns"`
    ARP   *ARPResult   `json:"arp"`
    MDNS  *MDNSResult  `json:"mdns"`
    HTTP  *HTTPResult  `json:"http"`
    {Name} *{Name}Result `json:"{name}"`  // add here
}

2b: Write prober function

Create /home/user/Strix/pkg/probe/{name}.go.

Rules:

• Pure function, no app/api imports
• Takes context.Context and ip string as first params
• Returns (*{Name}Result, error)
• Respects context deadline (timeout comes from runProbe)
• Returns nil, nil when device doesn't support this (NOT an error)
• Keep it simple -- one file, one function

Pattern:

package probe

import "context"

func Probe{Name}(ctx context.Context, ip string) (*{Name}Result, error) {
    // respect context deadline
    deadline, ok := ctx.Deadline()
    if !ok {
        // set sensible default
    }

    // do the probe work...

    // not supported = nil, nil (not an error)
    // found = &{Name}Result{...}, nil
    // actual error = nil, err
}

2c: Wire prober into runProbe

Edit /home/user/Strix/internal/probe/probe.go, add to runProbe() alongside other probers:

run(func() {
    r, _ := probe.Probe{Name}(ctx, ip)
    mu.Lock()
    resp.Probes.{Name} = r
    mu.Unlock()
})

All probers run in parallel inside the same run() pattern. The mutex protects writes to resp.Probes.

---

STEP 3: Add detector function

Edit /home/user/Strix/internal/probe/probe.go, add detector in Init():

// {Name} detector
detectors = append(detectors, func(r *probe.Response) string {
    // check probe results to determine device type
    // return type string or "" to pass
    if r.Probes.{Something} != nil && {condition} {
        return "{type_name}"
    }
    return ""
})

Detector rules

1. Return a SHORT type string: "homekit", "onvif", "tapo", etc.
2. Return "" (empty) to pass to the next detector
3. Detectors run in order -- put more specific detectors BEFORE generic ones
4. A detector can use ANY combination of probe results (ports, HTTP, ARP, mDNS, custom)
5. Don't do network I/O in detectors -- all data should come from probers

Type string convention

The type string is used by the frontend to select UI flow:

• "unreachable" -- device not found (set automatically, don't return this)
• "standard" -- default, normal camera (set automatically if no detector matches)
• "homekit" -- Apple HomeKit device
• Custom types: lowercase, one word, matches the protocol/brand name

---

STEP 4: Build and test

cd /home/user/Strix
go build ./...

If it compiles, rebuild Docker and test:

docker build -t strix:test .
docker rm -f strix
docker run -d --name strix --network host --restart unless-stopped strix:test
sleep 2

# test probe on a known device
curl -s "http://localhost:4567/api/probe?ip={DEVICE_IP}" | python3 -m json.tool

Verify:

1. New probe data appears in probes object (if new prober added)
2. type field correctly identifies the device
3. No errors in docker logs strix

---

STEP 5: Commit and push

cd /home/user/Strix
git add -A
git commit -m "Add {name} probe detector"
git push origin develop

---

CODE STYLE

pkg/probe/ files

• One file per prober
• Pure functions, no globals, no app imports
• context.Context as first param for anything with I/O
• Return nil, nil for "not applicable" (not an error)
• Short names: conn, resp, buf

internal/probe/probe.go

• Detectors are inline anonymous functions in Init()
• Keep detector logic minimal -- just check fields and return type
• If detector logic is complex (>10 lines), extract to a named function in the same file

models.go

• All result structs in one file
• JSON tags use lowercase with underscores
• Optional fields use omitempty
• Pointer types for probe results (nil = not collected)