selfphoneng/scripts/e2e/README.md

Browser checks

Dependency-free scripts that drive a headless Chromium (Edge/Chrome) over the DevTools protocol. They need an api serving the built web UI and a fresh database.

pnpm --filter @selfphoneng/web build && mkdir -p /tmp/sp && cp -r apps/web/dist /tmp/sp/public
# fresh DB, then from /tmp/sp:
DATABASE_URL=postgres://… INITIAL_ADMIN_EMAIL=admin@example.com INITIAL_ADMIN_PASSWORD=initial-admin-pw \
  node --env-file=<repo>/.env --conditions=source <repo>/apps/api/src/index.ts
BROWSER=/usr/bin/chromium node scripts/e2e/m1-t4.mjs http://127.0.0.1:8080 admin@example.com initial-admin-pw /tmp/shots
node scripts/e2e/mobile.mjs http://127.0.0.1:8080 admin@example.com correct-horse-battery-staple /tmp/shots

m1-t4.mjs is the M1-T4 acceptance flow: forced password change, then create/edit/delete a user.

M2 live loop (any Docker host, no host networking)

m2-live.mjs runs the whole chain on a private Docker network: a test Asterisk with AMI published on 127.0.0.1:15038, the api writing config into it, two baresip phones, and the dashboard in a browser. It checks that a new extension is live in Asterisk within 2 s, both phones show online, and the dots turn red during a call.

node scripts/e2e/m2-live.mjs up /tmp/sp-live-conf
# start the api with AMI_PORT=15038 AMI_SECRET=m2-live-secret ASTERISK_CONFIG_DIR=/tmp/sp-live-conf
node scripts/e2e/m1-t4.mjs http://127.0.0.1:8080 admin@example.com initial-admin-pw /tmp/shots
node scripts/e2e/m2-live.mjs check http://127.0.0.1:8080 admin@example.com correct-horse-battery-staple /tmp/shots
node scripts/e2e/m2-live.mjs down

m3-t1.mjs runs on the same setup (after m1-t4.mjs): it starts a fake voip.ms (a second Asterisk that accepts one sub-account) on the m2-live network, creates the trunk in the UI and waits for Registered, then presses Test. m3-t2.mjs (after m3-t1.mjs) adds a phone number routed to a user in the UI and checks sp-inbound in Asterisk, and that the extension can no longer be deleted. m3-t3.mjs (after both) checks the default outbound rules, the test box, a user's calling permission, and that reordering changes the include order in Asterisk. m3-t4.mjs (after all three) runs a tiny SMTP sink on 127.0.0.1:2525 and a baresip phone as 101, then checks the dashboard warnings, Settings → Email and Emergency, and the Test notifications button (email arrives, 101 rings, red banner). Nothing dials 911. The fake provider from m3-t1.mjs stays up until m2-live.mjs down: a container that took over its IP would be identified as the trunk.

pnpm test:asterisk (golden-file load test, call-test.ts with 30 s two-way audio between two baresip phones, and trunk-test.ts with registration against a fake provider) needs no api at all.

m3-t5.mjs is the M3-T5 acceptance and needs its own fresh database: it signs in with the initial password, walks all ten wizard steps against a fake voip.ms, then proves an inbound call to the DID rings the new user's softphone and an outbound call leaves with the DID as caller ID.

m4-t2.mjs is the M4-T2 acceptance (no browser): start m2-live with a voicemail dir (m2-live.mjs up <confDir> <vmDir>) and the api with VOICEMAIL_DIR=<vmDir>. It runs the SMTP sink (smtp-sink.mjs, shared with m3-t4.mjs), adds a user and extension 150 with voicemail, leaves a message from a Local channel, and checks the email arrives within 10 s with a PCM 8 kHz wav attached. The voicemail itself (MWI on a phone, *97) is covered by pnpm test:asterisk.

m4-t3.mjs <baseUrl> <adminEmail> <adminPassword> <vmDir> <shotsDir> is the M4-T3 acceptance on the same setup: it adds a user with extension 160 and a baresip phone with the mwi module, leaves messages, then as that user in the browser plays one (marks it read) and deletes another, each time checking the phone's MWI goes off within 5 s. It records the "no answer" greeting with a fake microphone and checks a caller hears it.

m5-t1.mjs <baseUrl> <adminEmail> <adminPassword> <shotsDir> is the M5-T1 check in Docker: start the api with WEBRTC_WS_URL=ws://127.0.0.1:18088/ws. m2-live.mjs up publishes Asterisk's HTTP server on 127.0.0.1:18088 and RTP 10000–10019/udp. The script sets LAN IP = the container and public IP = 127.0.0.1 (so ICE offers 127.0.0.1), restarts telephony, then signs in as a new user, enables the softphone with a fake microphone, and checks two-way audio to a baresip phone (171) in both directions.

m5-t2.mjs (after m5-t1.mjs, same user) starts two baresip phones (172, 173) through phones.mjs and checks the softphone features: the incoming call dialog, DTMF, hold (music on hold at the other end), call waiting, blind and attended transfer, and "Use here" moving the softphone to a second tab (cdp.mjs newTab()).

m5-t3.mjs (after m5-t2.mjs) is the M5-T3 acceptance. It watches People in one browser while 173 calls 172 and measures the time from Asterisk's hint changing to the dot changing (must be < 1 s). Then a second browser, as a new user with desk phone 174 and the softphone off, clicks Call on 172: 174 must ring first (Asterisk channel order), then the two are connected.

m6-t1.mjs needs m2-live started with a sounds dir (m2-live.mjs up <confDir> <vmDir> <soundsDir>) and the api with SOUNDS_DIR=<soundsDir>. It drops a generated wav on Sounds & music, uploads it, checks that Asterisk can play custom/<id>, renames it, and deletes it, which is refused while it's the recording notice. cdp.mjs fill()/click() take an optional selector to search within.

m6-t2.mjs (after m1-t4.mjs) creates ring group 500 in the UI with members 190 (never answers) and 191 (auto-answer), and checks that a call from 192 rings both and is answered by 191. It then switches the group to "one at a time" with 5 s each and checks that 190 rings alone before 191.

m6-t3.mjs (sounds dir like m6-t1.mjs) builds digital receptionist 600 in the UI. It uploads the greeting through the form's SoundPicker, sends key 1 to 193 and key 2 to 194, and then a Local channel sending DTMF presses 1, 2 and a wrong key (5): 193 or 194 rings, or "not a valid choice" plays and then the greeting again.

m6-t4.mjs builds office hours in the UI. It paints Saturday 10–11 on the grid and checks that Force closed / Force open route a call to 196 / 195 and set or clear AstDB SP/TCOVR/<id>. *321 and *331 from the admin's own phone (197) must reach the database and the page, *311 from a non-admin phone must get "no service", and the 2026 US federal holiday import must reach the dialplan.

m6-t7.mjs (after m1-t4.mjs) checks the feature codes from a real phone (198, belonging to user fwd@example.com):

• *78/*79 set DND (the database, AstDB and the page follow, and calls meanwhile go to voicemail).
• *72199/*73 forward everything to 199. A *72 to an outside number this user may not call is refused, removed from AstDB, and shown as a toast.
• /me/settings saves a no-answer forward.
• *43 echo test and *65 say my extension.
• **200 and *8 pick up a call ringing at 200.

m6-check.mjs is the M6 milestone check (after m1-t4.mjs). It builds DID → office hours → menu (1 → a ring group of 221 and 222, 2 → 223; closed → 220's voicemail) and calls the trunk entry with a caller ID over AMI. It checks the closed path; *31N from an admin's phone (220) forcing open; the menu keys; and a number blocked on Blocked numbers being rejected (cause 21) while other callers still get through.

m7-t2.mjs (after m1-t4.mjs, with the Asterisk image rebuilt for cdr_manager.conf) makes real calls: 231 → 230 answered, then 231 → 230 while 230 is on DND (voicemail). It waits for them to reach calls through the AMI Cdr events. It then checks the admin Call history (status filter, expanded legs, CSV export), and as user hist@example.com (230): the missed call on Home, /me/calls, and Call back, which rings 230's desk phone first and then 231.

m7-t3.mjs needs m2-live with sounds and recordings dirs (m2-live.mjs up <conf> <vm> <sounds> <recs>) and the api with SOUNDS_DIR and RECORDINGS_DIR. It picks the recording notice under Settings → Recording and ticks "Record every call" on a phone number in the UI. A caller then reaches 240 (AMI Originate into the trunk entry, ami.mjs): the notice plays, MixMonitor records, and the call's player in Call history serves the wav. The recording is also on the Recordings page.

m7-t4.mjs: seven auto-answer phones (260-266) and three concurrent calls (channel originate). The dashboard's "Calls now" list must show exactly the channels in core show channels concise. Then the admin hangs up one call (confirm dialog) and transfers another's caller to 266; the list matches Asterisk after each step.

m8-t1.mjs: three auto-answer phones (agents 280 and 281, overflow 282). It builds queue 550 in the UI (wait 20 s, then 282; callers wait even with no agent logged in), pauses 280 with *45550 from the phone and 281 with the queue page's button, and puts three callers in line (AMI Originate of Local/550@sp-from-ext). The queue page and the dashboard tile show them live; unpausing 281 connects the longest waiting caller, a second caller goes to 282 after 20 s, a third hangs up. *45550 logs 280 back in, *45551 (not an agent) plays ss-noservice, and the stats count 3 calls: 1 answered, 1 hung up, 1 timed out.

m8-t2.mjs: three auto-answer phones (290 parks, 291 belongs to a user, 292 picks up). A caller reaches 290, which blind-transfers it to 700 (AMI BlindTransfer, what a phone's transfer key does): it lands on 701 and the slot's hint goes InUse. The dashboard and the user's Home list it, the live call shows "Parked", and Pick up 701 on Home rings 291, which dials 701 and talks to the caller. A second park takes 702 (findslot = next) and 292 picks it up by dialing 702.

m8-t3.mjs needs m2-live with a recordings dir. It builds room 800 in the UI (PIN 4826, admin PIN 917355, at most 3, recorded; the form refuses an admin PIN equal to the PIN). Callers are Local channels typing with SendDTMF: a wrong PIN plays conf-invalidpin and asks again, the admin PIN joins with flags AM in confbridge list, a 4th caller is turned away, and the recordings appear on the Recordings page. Room 801 has no PIN; desk phones 300 and 301 dial in.

m8-t4.mjs: phones 320 and 321 answer manually but honor the auto-answer headers (sip_autoanswer=yes, like a Yealink or Grandstream); 322 auto-answers and does the paging. It builds paging group 850 in the UI. An ordinary call to 320 keeps ringing; a page from 322 is answered by 320 and 321 by themselves (the INVITE carries Call-Info: …;answer-after=0 and Alert-Info, seen with pjsip set logger on), muted (m in confbridge list), and 322 (busy, a member too) isn't paged. *80320 is two-way; *80999 plays ss-noservice.

m8-t5.mjs needs m2-live with a moh dir (m2-live.mjs up 5th argument, the api with MOH_DIR). It makes the set sales-jazz on Sounds & music (refusing a bad name), uploads a tone into it, and checks moh show files, a queue caller hearing it ("Started music on hold, class 'sales-jazz'"), a user's moh_suggest, delete refused while used, and the default music once the set is empty.

m8-check.mjs is the M8 milestone check (after m1-t4.mjs): phones 340–343, a DID to queue 552 (550 is m8-t1's) (agents 340 and 341, 30 s, then 342). 340 pauses with *45552 and 341 from the api, three callers come in from the trunk entry and wait; 341 comes back, gets caller 1, parks them on 701 (AMI BlindTransfer to 700), 343 dials 701 and picks up; 341 takes caller 2 and caller 3 rings 342 after 30 s. The stats count 3 offered, 2 answered, 1 timed out.

m9-t1.mjs starts a fake voip.ms API on 127.0.0.1:18090 (start the api with VOIPMS_API_URL=http://127.0.0.1:18090/api/v1/rest.php; the optional 5th argument is the api log). With a wrong API password the trunk form shows voip.ms's reason and the built-in server list; with the right one the list comes from getServersInfo. Import numbers from voip.ms imports one of two numbers with its description as the label, the list marks it added, Phone numbers shows it, and the api log never contains the API password. The trunk is created turned off, so nothing registers to the real voip.ms.

m10-t1.mjs (after m1-t4.mjs) adds a Yealink on Phones (a short MAC is refused), reads the URL, user name and password from the drawer, and checks the preview masks the passwords. Then, as the phone: /prov/<mac>.cfg answers 401 without credentials and the file with them, and a baresip phone given only the account from that file registers. The list then shows when it fetched its settings.

m10-t2.mjs (after m1-t4.mjs) gives 360 BLF keys for 361 and the parking slot 701 and a Yealink. From the phone's side: the config names the keys, clears the rest, has Central time, and its phonebook URL (with the phone's credentials in it) returns the extensions. Asterisk has hints for 361 and 701. With the company time zone set to Asia/Tokyo, the Phones page warns that phones show UTC (the zone is put back after).

m10-t3.mjs (after m1-t4.mjs) as user lin@example.com (370): My settings → Set up a phone app shows the Linphone QR and its link. Fetching the link with no other credentials (as Linphone does) returns the account XML; a changed token gets 404; and a baresip phone given only the user name and password from the XML registers. Scanning the QR with the Linphone app is left to a person.

m10-t4.mjs (after m1-t4.mjs): 380 calls 381; the admin, whose extension is 382 (desk phone), presses Listen, Whisper and Barge on the dashboard. Each time 382 rings, answers and runs ChanSpy(PJSIP/380-…, qE / qEw / qEB); hanging up the call ends the barge leg too.

m11-t1.mjs (after m1-t4.mjs; needs the Asterisk image with sp-vm-pincheck; 5th argument: the generated config dir) clicks through admin pages and fails on any Content-Security-Policy violation; turns the phone ACL on with the phones' Docker network unlisted (390 can't register: "Not match Endpoint ACL") and then listed (it registers); and changes 391's voicemail PIN from the phone with *98 and DTMF: 1111 is refused by sp-vm-pincheck, 7391 is stored by the api and survives a later config write.

m11-t1-totp.mjs (after m1-t4.mjs): the admin turns two-step sign-in on from My account (a wrong code refused, 10 recovery codes shown), signs out and back in with password then a code, then with a recovery code (9 left), and turns it off with the password. Codes come from apps/api/src/auth/totp.ts (node runs it directly).

m11-t2.mjs <baseUrl> <adminEmail> <adminPassword> <tlsDir> <shotsDir> (after m1-t4.mjs; start the api with TLS_DIR=<tlsDir>, an empty folder) makes a throwaway CA with the image's openssl and starts a fake voip.ms that only speaks TLS (sp-live-tlsprov, certificate for its IP). Settings → Security first says there is no certificate, then shows the PBX's; the admin turns TLS on and telephony restarts. A TLS + SRTP trunk must not register while the CA is untrusted (verify_server logs "certificate is untrusted"), then registers once it is; a call carries RTP/SAVP with an a=crypto key and audio both ways; a renewed PBX certificate is served after module reload res_pjsip.so (what scripts/tls-cert.sh runs) while that call stays up.

m11-t3.mjs (no api, no m2-live; needs the images from docker compose build) is the M11-T3 acceptance on two throwaway servers: privileged docker:28-dind containers, each running the real compose stack with the images loaded from this host. Server A gets data (a new admin password, user Dan with 201, a trunk with a password, voicemail, a sound, music on hold, a recording) and Back up now (0600, no generated config, no recordings, the trunk password only encrypted in db.sql). Server B, a fresh install with the same .env, refuses the backup under another APP_SECRET and on a "no", then restores it: the admin signs in with A's password, 201 and the trunk are back, the trunk password decrypts into Asterisk's config, the files match byte for byte. KEEP=1 leaves both servers running.

m11-t4.mjs (m2-live started with a voicemail dir, after m1-t4.mjs; stop your api first: the script starts and restarts it, with the api's environment taken from its own) is the M11-T4 acceptance, with mail going to a local SMTP sink on 127.0.0.1:2525. Settings → Network turns on the public IP watch and Settings → Recording keeps voicemail 30 days. A real voicemail, made 40 days old in the database, is removed through Asterisk (files gone) when the api restarts. The watch finds the real public IP (needs internet), saves it, asks for a restart and emails the admins. A trunk to 192.0.2.1 (nothing answers) gets exactly one "is down" email after 5 minutes, and a disk under 10% free gets a "Disk almost full" email. About 8 minutes.

m11-t5.mjs (m2-live with a voicemail dir, after m1-t4.mjs) is the M11-T5 acceptance. An API token made in Settings → API tokens adds extension 201 from a script (no cookie, no CSRF header), can't manage tokens, and is refused once revoked in the browser. A webhook added in Settings → Webhooks to a receiver on this host is refused as a private address until that's allowed, then the test arrives signed. A real call to 201 (Local channel, voicemail answers) delivers call.started, call.ended and voicemail.new, all with valid signatures; the receiver fails the first one with a 500 and it comes again a minute later. About 2 minutes.

m11-t6.mjs (m2-live with a voicemail dir, after m1-t4.mjs) is the M11-T6 acceptance. Three real calls: one answered by phone 260, one to 261's voicemail and one through queue 560 that 260 answers. /api/reports/summary agrees with call history (today's bars, missed, this hour in the heatmap) and /api/reports/queues shows 560 with 1 offered, 1 answered, 100% in time. The Reports page shows the same at 1280 px and 390 px, without scrolling sideways on the phone. About 3 minutes.

m11-t7.mjs (m2-live, after m1-t4.mjs) checks the admin guide (docs/admin-guide/). A baresip phone signs in to extension 270 with the wrong password; fail2ban's stock asterisk filter (fail2ban-regex, Debian's fail2ban) matches the resulting security.log lines with the phone's IP, and CrowdSec's crowdsecurity/asterisk collection parses them (cscli explain). It then takes the guide's screenshots into the given folder (run it with docs/admin-guide/img) and checks every relative link, image and #anchor in the guide. About 4 minutes.