prospecting-workflow

name	prospecting-workflow
description	End-to-end guided workflow — build a Saber company list, run research signals across every company in parallel, find named contacts, and export a combined CSV to the desktop. Walks the user through each stage with confirmations and a credit-discipline guard. Use when you want one signed-off CSV instead of running build-account-list, create-company-signals, and build-contact-list separately.

Prospecting Workflow

A guided workflow for:

1. Building a filtered company list with a live count preview
2. Running research signals against every company in the list
3. Searching for contacts at those companies
4. Exporting a combined CSV (companies + signals + contacts) to ~/Desktop/

Uses the Saber CLI (/opt/homebrew/bin/saber). Auth is handled automatically via stored credentials — no API key needed.

---

Trigger Discipline — Read This First

saber subscription trigger has no idempotency and no cache. Every call spawns a fresh batch of signal runs across every company in the list and charges credits for every one — including companies already answered by a prior trigger. Re-triggering is how credit budgets get destroyed.

Rule: Never call subscription trigger on a subscription whose lastRunAt is non-null, or that already has signals in saber signal list --subscription-id <id>. Use the pre-trigger guard in Stage 3. No exceptions — not for stalled signals, not for "just to be safe", not for retries.

A second, equally important rule: never call saber subscription start in this workflow. Subscriptions are created with --run-once, which is what keeps them as one-off jobs (no Temporal schedule, cron field dormant). Calling start materializes the Temporal schedule from the cron expression and converts every "one-off" into a recurring monthly job that will fire on its own at the cron's next tick. There is no DELETE endpoint for company signal subscriptions, so a started subscription must be stopped to disarm — and any drift to status=active is the failure mode that drains credits in bulk.

Safe vs guarded commands:

Command	Type	Can you call it freely?
saber signal list --subscription-id <id>	read-only	Yes — poll as often as you like
saber signal get <signalId>	read-only	Yes
saber subscription get <id>	read-only	Yes
saber subscription list	read-only	Yes
saber list company companies <listId>	read-only	Yes
saber list company count-preview	read-only	Yes
saber subscription stop <id>	safe write	Yes — disarms a runaway active subscription
saber subscription trigger <id>	guarded write	Only via the pre-trigger guard — aborts if already run
saber subscription start <id>	forbidden in this workflow	Never. Converts one-off subs into recurring monthly jobs.

Progress polling during a running batch uses only read-only commands. It never interacts with trigger and never causes re-runs. Poll freely.

Session resumption: If you pick up a conversation that already has a subscription ID (from prior context, a passed-in ID, or anything other than a subscription you just created in the current session), assume it has already been triggered. Run the guard before any trigger call — never assume "fresh". Also verify status=stopped before doing anything else; if it's active, stop it immediately.

---

Stage 1: Filter Setup & Preview

Ask the user for filters. Collect the following:

Industry (required) Must be an exact string from the list below. Validation is strict and case-sensitive. If the user's input doesn't exactly match, find the closest valid option, show it to them, and confirm before proceeding. Multiple industries are supported (repeatable flag).

Common pitfalls:

• marketing & advertising is NOT valid → use advertising services or marketing services
• hospitals & health care is NOT valid → use hospitals and health care
• information technology & services IS valid (ampersand kept)
• software development IS valid

Full valid industry list: abrasives and nonmetallic minerals manufacturing, accessible architecture and design, accessible hardware manufacturing, accommodation and food services, accounting, administration of justice, administrative and support services, advertising services, agricultural chemical manufacturing, agriculture, construction, mining machinery manufacturing, air, water, and waste program management, airlines and aviation, alternative dispute resolution, alternative fuel vehicle manufacturing, alternative medicine, ambulance services, amusement parks and arcades, animal feed manufacturing, animation, animation and post-production, apparel & fashion, apparel manufacturing, appliances, electrical, and electronics manufacturing, architectural and structural metal manufacturing, architecture and planning, armed forces, artificial rubber and synthetic fiber manufacturing, artists and writers, arts & crafts, assurances, audio and video equipment manufacturing, automation machinery manufacturing, automotive, aviation & aerospace, aviation and aerospace component manufacturing, baked goods manufacturing, banking, bars, taverns, and nightclubs, bed-and-breakfasts, hostels, homestays, beverage manufacturing, biomass electric power generation, biotechnology, biotechnology research, blockchain services, blogs, boilers, tanks, and shipping container manufacturing, book and periodical publishing, book publishing, breweries, broadcast media production and distribution, building construction, building equipment contractors, building finishing contractors, building materials, building structure and exterior contractors, business consulting and services, business content, business intelligence platforms, business supplies & equipment, cable and satellite programming, capital markets, caterers, chemical manufacturing, chemical raw materials manufacturing, child day care services, chiropractors, circuses and magic shows, civic and social organizations, civil engineering, claims adjusting, actuarial services, clay and refractory products manufacturing, climate data and analytics, climate technology product manufacturing, coal mining, collection agencies, commercial and industrial equipment rental, commercial and industrial machinery maintenance, commercial and service industry machinery manufacturing, commercial real estate, communications equipment manufacturing, community development and urban planning, community services, computer and network security, computer games, computer hardware, computer hardware manufacturing, computer networking, computer networking products, computers and electronics manufacturing, conservation programs, construction, construction hardware manufacturing, consumer electronics, consumer goods, consumer goods rental, consumer services, correctional institutions, cosmetics, cosmetology and barber schools, courts of law, credit intermediation, cutlery and handtool manufacturing, dairy, dairy product manufacturing, dance companies, data infrastructure and analytics, data security software products, defense & space, defense and space manufacturing, dentists, design, design services, desktop computing software products, digital accessibility services, distilleries, e-learning, e-learning providers, economic programs, education, education administration programs, education management, electric lighting equipment manufacturing, electric power generation, electric power transmission, control, and distribution, electrical equipment manufacturing, electronic and precision equipment maintenance, embedded software products, emergency and relief services, engineering services, engines and power transmission equipment manufacturing, entertainment, entertainment providers, environmental quality programs, environmental services, equipment rental services, events services, executive offices, executive search services, fabricated metal products, fabrication de véhicules à moteur, facilities services, family planning centers, farming, farming, ranching, forestry, fashion accessories manufacturing, financial services, fine art, fine arts schools, fire protection, fisheries, flight training, food & beverages, food and beverage manufacturing, food and beverage retail, food and beverage services, food production, footwear and leather goods repair, footwear manufacturing, forestry and logging, fossil fuel electric power generation, freight and package transportation, fruit and vegetable preserves manufacturing, fuel cell manufacturing, fundraising, funds and trusts, furniture, furniture and home furnishings manufacturing, gambling facilities and casinos, geothermal electric power generation, glass product manufacturing, glass, ceramics and concrete manufacturing, golf courses and country clubs, government administration, government relations, government relations services, graphic design, ground passenger transportation, health and human services, health, wellness & fitness, higher education, highway, street, and bridge construction, historical sites, holding companies, home health care services, horticulture, hospitality, hospitals, hospitals and health care, hotels and motels, household and institutional furniture manufacturing, household appliance manufacturing, household services, housing and community development, housing programs, human resources, human resources services, hvac and refrigeration equipment manufacturing, hydroelectric power generation, import & export, individual and family services, industrial automation, industrial machinery manufacturing, industry associations, information services, information technology & services, insurance, insurance agencies and brokerages, insurance and employee benefit funds, insurance carriers, interior design, international affairs, international trade and development, internet marketplace platforms, internet news, internet publishing, interurban and rural bus services, investment advice, investment banking, investment management, it services and it consulting, it system custom software development, it system data services, it system design services, it system installation and disposal, it system operations and maintenance, it system testing and evaluation, it system training and support, janitorial services, landscaping services, language schools, laundry and drycleaning services, law enforcement, law practice, leasing non-residential real estate, leasing residential real estate, leather product manufacturing, legal services, legislative offices, leisure, travel & tourism, libraries, lime and gypsum products manufacturing, loan brokers, luxury goods & jewelry, machinery manufacturing, magnetic and optical media manufacturing, manufacturing, maritime, maritime transportation, market research, marketing services, mattress and blinds manufacturing, measuring and control instrument manufacturing, meat products manufacturing, mechanical or industrial engineering, media and telecommunications, media production, medical and diagnostic laboratories, medical device, medical equipment manufacturing, medical practices, mental health care, metal ore mining, metal treatments, metal valve, ball, and roller manufacturing, metalworking machinery manufacturing, military and international affairs, mining, mobile computing software products, mobile food services, mobile gaming apps, motor vehicle manufacturing, motor vehicle parts manufacturing, movies and sound recording, movies, videos, and sound, museums, museums, historical sites, and zoos, music, musicians, nanotechnology research, natural gas distribution, natural gas extraction, newspaper publishing, non-profit organization management, non-profit organizations, nonmetallic mineral mining, nonresidential building construction, nuclear electric power generation, nursing homes and residential care facilities, office administration, office furniture and fixtures manufacturing, oil and coal product manufacturing, oil and gas, oil extraction, oil, gas, and mining, online and mail order retail, online audio and video media, online media, operations consulting, optometrists, outpatient care centers, outsourcing and offshoring consulting, outsourcing/offshoring, packaging & containers, packaging and containers manufacturing, paint, coating, and adhesive manufacturing, paper & forest products, paper and forest product manufacturing, pension funds, performing arts, performing arts and spectator sports, periodical publishing, personal and laundry services, personal care product manufacturing, personal care services, pet services, pharmaceutical manufacturing, philanthropic fundraising services, philanthropy, photography, physical, occupational and speech therapists, physicians, pipeline transportation, plastics and rubber product manufacturing, plastics manufacturing, political organizations, postal services, primary and secondary education, primary metal manufacturing, printing services, professional organizations, professional services, professional training and coaching, program development, public assistance programs, public health, public policy, public policy offices, public relations and communications services, public safety, racetracks, radio and television broadcasting, rail transportation, railroad equipment manufacturing, ranching, ranching and fisheries, real estate, real estate agents and brokers, real estate and equipment rental services, recreational facilities, regenerative design, religious institutions, renewable energy equipment manufacturing, renewable energy power generation, renewable energy semiconductor manufacturing, renewables & environment, repair and maintenance, research, research services, residential building construction, restaurants, retail, retail apparel and fashion, retail appliances, electrical, and electronic equipment, retail art dealers, retail art supplies, retail books and printed news, retail building materials and garden equipment, retail florists, retail furniture and home furnishings, retail gasoline, retail groceries, retail health and personal care products, retail luxury goods and jewelry, retail motor vehicles, retail musical instruments, retail office equipment, retail office supplies and gifts, retail pharmacies, retail recyclable materials & used merchandise, reupholstery and furniture repair, robot manufacturing, robotics engineering, rubber products manufacturing, satellite telecommunications, savings institutions, school and employee bus services, seafood product manufacturing, secretarial schools, securities and commodity exchanges, security and investigations, security guards and patrol services, security systems services, seguros, semiconductor manufacturing, semiconductors, services for renewable energy, services for the elderly and disabled, sheet music publishing, shipbuilding, shuttles and special needs transportation services, sightseeing transportation, skiing facilities, smart meter manufacturing, soap and cleaning product manufacturing, social networking platforms, software development, solar electric power generation, sound recording, space research and technology, specialty trade contractors, spectator sports, sporting goods, sporting goods manufacturing, sports and recreation instruction, sports teams and clubs, spring and wire product manufacturing, staffing and recruiting, steam and air-conditioning supply, strategic management services, subdivision of land, sugar and confectionery product manufacturing, surveying and mapping services, taxi and limousine services, technical and vocational training, technology, information and internet, technology, information and media, telecommunications, telecommunications carriers, telephone call centers, temporary help services, textile manufacturing, theater companies, think tanks, tobacco, tobacco manufacturing, translation and localization, transportation equipment manufacturing, transportation programs, transportation, logistics, supply chain and storage, transportation/trucking/railroad, travel arrangements, truck transportation, trusts and estates, turned products and fastener manufacturing, urban transit services, utilities, utilities administration, utility system construction, vehicle repair and maintenance, venture capital and private equity principals, veterinary, veterinary services, vocational rehabilitation services, warehousing, warehousing and storage, waste collection, waste treatment and disposal, water supply and irrigation systems, water, waste, steam, and air conditioning services, wellness and fitness services, wholesale, wholesale alcoholic beverages, wholesale apparel and sewing supplies, wholesale appliances, electrical, and electronics, wholesale building materials, wholesale chemical and allied products, wholesale computer equipment, wholesale drugs and sundries, wholesale food and beverage, wholesale footwear, wholesale furniture and home furnishings, wholesale hardware, plumbing, heating equipment, wholesale import and export, wholesale luxury goods and jewelry, wholesale machinery, wholesale metals and minerals, wholesale motor vehicles and parts, wholesale paper products, wholesale petroleum and petroleum products, wholesale photography equipment and supplies, wholesale raw farm products, wholesale recyclable materials, wind electric power generation, wine & spirits, wineries, wireless services, wood product manufacturing, writing and editing, zoos and botanical gardens

Employee size (optional, repeatable) Must be one of these exact strings: 1-10, 11-50, 51-200, 201-500, 501-1K, 1K-5K, 5K-10K, 10K+

If the user says "over 100 employees", explain that the closest options are 51-200 (includes 51–100) or starting from 201-500. Confirm which ranges they want.

Country (optional, repeatable) ISO 2-letter codes: e.g. NL, US, GB, DE, FR

Technology (optional, repeatable) Technology slugs: e.g. stripe, hubspot, salesforce, intercom

Once filters are collected, run a count preview:

saber list company count-preview \
  --industry "<industry>" \
  [--size "<range>" --size "<range>"] \
  [--country "<code>"] \
  [--technology "<slug>"] \
  --json

Show the user: "X companies matched. Each signal costs 2 credits — running N signals against X companies will cost N×X×2 credits total."

For example, 3 signals against 2,466 companies = 7,398 signal runs × 2 credits = 14,796 credits.

Ask: "Happy with this count, or would you like to adjust the filters?"

Iterate until the user confirms. Do not recommend narrowing filters based on list size — that's the user's call. Just be transparent about the credit implications.

---

Stage 2: Create the List

Ask the user for a list name, or suggest one based on the filters (e.g. "NL SaaS 200+").

saber list company create \
  --name "<name>" \
  --industry "<industry>" \
  [--size "<range>" --size "<range>"] \
  [--country "<code>"] \
  [--technology "<slug>"] \
  --json

Save the returned list ID — you'll need it for every subsequent step.

Confirm to the user: "List '[name]' created (ID: <listId>)."

---

Stage 3: Signal Definition

Ask the user:

"What signals would you like to run against these companies? Just list the questions — I'll handle the answer types."

Infer the answer type from the question. Do not ask the user to specify it. Use this guide:

Answer type	When to use	Example question
boolean	Pure yes/no with no follow-up detail needed	"Is this company B2B?", "Do they have a mobile app?"
number	Numeric values	"How many employees?", "What is their Trustpilot rating?"
open_text	Free-form descriptions or summaries (single field, no structure)	"Describe what they sell", "What is their main product?"
list	Multiple distinct items of the same type (no boolean flag needed)	"What are their main competitors?", "What markets do they serve?"
url	A single URL	"What is their LinkedIn URL?", "What is their careers page?"
percentage	A percentage value	"What percentage of revenue is from SaaS?"
currency	A monetary amount	"What is their estimated ARR?", "What was their last funding amount?"
json_schema	Compound questions (yes/no + follow-up detail), or any question requiring multiple structured fields	"Have they raised funding? If so, what stage and amount?", "Are they hiring? If so, which roles?", "Have they launched new features? If so, which?"

When to prefer json_schema: Any question that contains "if so", "if yes", or implicitly combines a boolean with follow-up detail fields should use json_schema. Design the schema to include a boolean flag (e.g. raised, expanding, hiring) plus the relevant detail fields (strings, arrays, etc.). This produces cleaner, more queryable output than open_text.

Examples of json_schema schemas for common compound questions:

• Funding: { raised: boolean, stage: string, amount: string, date: string }
• Market expansion: { expanding: boolean, markets: [string] }
• New features: { introduced: boolean, features: [string] }
• Events: { attending: boolean, events: [string] }
• Hiring: { hiring: boolean, roles: [string] }

For json_schema: the CLI does not support passing an output schema. Use the Saber API directly instead:

# Test signal (sync)
curl -s -X POST 'https://api.saber.app/v1/companies/signals/sync' \
  -H 'Content-Type: application/json' \
  -H "Authorization: Bearer $SABER_API_KEY" \
  -d '{
    "domain": "<domain>",
    "question": "<question>",
    "answerType": "json_schema",
    "outputSchema": {
      "type": "object",
      "properties": {
        "field1": { "type": "string" },
        "field2": { "type": "array", "items": { "type": "string" } }
      }
    }
  }'

# Template creation (for subscriptions)
curl -s -X POST 'https://api.saber.app/v1/companies/signals/templates' \
  -H 'Content-Type: application/json' \
  -H "Authorization: Bearer $SABER_API_KEY" \
  -d '{
    "name": "<template name>",
    "question": "<question>",
    "answerType": "json_schema",
    "outputSchema": { ... }
  }'

The json_schema path requires SABER_API_KEY to be set in the environment for curl. Skip it (use open_text instead) if the user can't supply a key — every other answer type works through the CLI without a key.

Present the inferred signal list back to the user — showing question + answer type — and ask:

"Before committing to the full run, would you like to test these signals against a sample of up to 5 companies? This lets you review answer quality and refine any questions before spending credits on the full list. (optional)"

If the user wants to test:

1. Fetch the company list and take the first 5 domains:

saber list company companies <listId> --json

Extract up to 5 domain values.

2. Launch all N×M signals simultaneously as background Bash tasks — do not run them sequentially. Fire every domain × signal combination at once:

saber signal --domain <domain1> --question "<q1>" --answer-type <type> --yes --json  # background
saber signal --domain <domain1> --question "<q2>" --answer-type <type> --yes --json  # background
saber signal --domain <domain2> --question "<q1>" --answer-type <type> --yes --json  # background
# ... all N×M in parallel

Tell the user: "Running [N×M] test signals in parallel — expect results in ~60–90 seconds."

Collect results as background tasks complete. Display a running table as answers arrive, then show the full table once all are done.

3. Ask the user: "How do the results look? Would you like to adjust any questions before running the full list?"

4. If they want to refine: update the signal questions, re-infer answer types, and re-run using the same parallel launch pattern. Revised question wording automatically bypasses the cache (cache key = domain + question), so no extra flags needed. Only use --force-refresh if re-running the exact same question text to get a fresher result. Repeat until the user is satisfied.

5. Once happy: "Ready to run [N] signals against [M] companies ([N×M] total = [N×M×2] credits)? This will start the full run."

If the user skips testing:

Ask directly: "Ready to run [N] signals against [M] companies ([N×M] total = [N×M×2] credits)? Confirm to proceed."

Wait for explicit confirmation before proceeding.

Once confirmed, for each signal:

1. Create a template:

saber template create \
  --name "<signal name>" \
  --question "<question text>" \
  --answer-type <type> \
  --json

Save each template ID.

2. Create a subscription (one per template) — one-off only:

saber subscription create \
  --list <listId> \
  --template <templateId> \
  --frequency monthly \
  --run-once \
  --yes \
  --json

Save each subscription ID.

--frequency monthly --run-once is required by the API (a frequency must be set), but --run-once instructs the backend to leave the subscription in status=stopped and never create a Temporal schedule entry. With no schedule entry, the cron field in the record is dormant data — it will not fire. Do not call saber subscription start on these subscriptions. Doing so creates the Temporal schedule from the cron field and converts a one-off into a recurring monthly job. (This is exactly the failure mode that has burned thousands of credits in past incidents — every "one-off" sub had been started and then fired again on the cron's next monthly tick.)

3. Trigger each subscription — via the pre-trigger guard:

Trigger through the guard without ever calling start. The guard MUST be used for every trigger call:

SUB_ID="<subscriptionId>"

# Pre-trigger guard — refuses if subscription has already run
LAST_RUN=$(saber subscription get "$SUB_ID" --json | jq -r '.lastRunAt // empty')
if [ -n "$LAST_RUN" ]; then
  echo "ABORT: $SUB_ID already ran at $LAST_RUN — re-trigger would re-charge every company in the list"
  exit 1
fi

SIGNAL_COUNT=$(saber signal list --subscription-id "$SUB_ID" --limit 1 --json | jq '.total')
if [ "$SIGNAL_COUNT" -gt 0 ]; then
  echo "ABORT: $SUB_ID already has $SIGNAL_COUNT signals spawned — re-trigger would duplicate"
  exit 1
fi

# One-off safeguard — refuse if the subscription is somehow active
STATUS=$(saber subscription get "$SUB_ID" --json | jq -r '.status')
if [ "$STATUS" != "stopped" ]; then
  echo "ABORT: $SUB_ID has status=$STATUS (expected stopped). A Temporal schedule is armed; this is no longer a one-off."
  echo "Investigate before triggering — run 'saber subscription stop $SUB_ID' to disarm, then re-evaluate."
  exit 1
fi

saber subscription trigger "$SUB_ID" --yes --verbose

Use --verbose on trigger to confirm HTTP 200 — that is the only confirmation needed. Do NOT re-trigger if the displayed subscription status still shows "stopped" after triggering; that's the expected end state for a one-off, and the status display also lags briefly behind. If you need to verify the trigger landed, call saber subscription get "$SUB_ID" and check lastRunAt — don't re-trigger.

4. Post-trigger safeguard — verify the subscription stayed one-off:

After triggering each subscription, confirm it is still status=stopped. A one-off should never end up active:

sleep 2
POST_STATUS=$(saber subscription get "$SUB_ID" --json | jq -r '.status')
if [ "$POST_STATUS" != "stopped" ]; then
  echo "WARNING: $SUB_ID is now status=$POST_STATUS after trigger. Disarming immediately to prevent monthly re-fire."
  saber subscription stop "$SUB_ID"
fi

If any subscription ever shows status=active, stop it immediately — that is the failure mode that drains credits on the cron's next tick.

Important: Subscription triggers always do a full fresh run — they do NOT use the signal cache (domain + question). The guard above is the only sanctioned way to call trigger. Bypassing it to recover stalled signals or retry a failed-looking trigger will re-run and re-charge credits for every company in the list. There is no DELETE endpoint for company signal subscriptions in the public API, so the only available defenses are --run-once at create time and stop if a subscription ever drifts to active.

Confirm: "[N] signals queued across [M] companies ([N×M] signal runs started)."

---

Stage 4: Time Estimate & Polling

Estimate completion time using this benchmark:

• Saber processes signals in parallel server-side — the bottleneck is backend concurrency, not list size
• Each individual signal takes ~30–60 seconds; the backend runs many simultaneously
• Use this table as your guide:

List size	Estimated time
< 100 companies	5–15 minutes
100–500 companies	10–30 minutes
500–2,000 companies	15–60 minutes
2,000+ companies	45–90 minutes

• Adding more signals per company has minimal impact on total time (they run in parallel per company too)
• Do NOT use N_companies × seconds — that assumes sequential execution and will overestimate by 10–100×

Tell the user the estimate and that you'll check status on request.

To check status, run this progress snapshot. All commands are read-only — they never cause re-runs or charge credits:

SUB_ID="<subscriptionId>"
LIST_ID="<listId>"   # from subscription or saved from Stage 2

# Expected total = number of companies in the list (one signal per company per subscription)
LIST_SIZE=$(saber list company companies "$LIST_ID" --limit 1 --json | jq '.total')

# Live counts by status — use --limit 1, only .total matters
PROC=$(saber signal list --subscription-id "$SUB_ID" --status processing --limit 1 --json | jq '.total')
DONE=$(saber signal list --subscription-id "$SUB_ID" --status completed   --limit 1 --json | jq '.total')
FAIL=$(saber signal list --subscription-id "$SUB_ID" --status failed      --limit 1 --json | jq '.total')

SPAWNED=$((PROC + DONE + FAIL))
REMAINING=$((LIST_SIZE - DONE - FAIL))
PCT=$((LIST_SIZE > 0 ? 100 * DONE / LIST_SIZE : 0))

echo "Progress: $DONE / $LIST_SIZE complete (${PCT}%)"
echo "  in-flight (processing): $PROC"
echo "  completed:              $DONE"
echo "  failed:                 $FAIL"
echo "  spawned so far:         $SPAWNED / $LIST_SIZE"
echo "  remaining to finish:    $REMAINING"

Key fields to always report to the user:

1. Completed — finished signals
2. Processing — in flight right now
3. Failed — gave up
4. Remaining — LIST_SIZE - (completed + failed) — the ones that still need to resolve
5. Percent complete — completed / list_size

Early in a run, SPAWNED < LIST_SIZE because Saber is still queuing signals server-side. Later, SPAWNED == LIST_SIZE and the only movement is processing → completed/failed.

Pagination note: Use --limit 1 for status counts — only the .total field matters, not the individual rows. To inspect which specific companies are stalled, paginate the processing list:

saber signal list --subscription-id "$SUB_ID" --status processing --json       # first 25
saber signal list --subscription-id "$SUB_ID" --status processing --offset 25 --json

To inspect a specific stalled signal:

saber signal get <signalId> --json

After each snapshot, report progress to the user and ask: "Still waiting on [N] signals. Check again?"

When remaining == 0: "All signals finished. Ready for contact search."

If any show "status": "failed", report which company + question failed.

If signals are genuinely stalled (processing for >5 minutes): Do NOT re-trigger the subscription. The pre-trigger guard will refuse it anyway, and bypassing the guard to "recover" stalled signals would re-charge credits for every company already answered. The only acceptable options are:

1. Wait — Saber's backend will eventually time out and mark the signal failed. Most stalls clear within 10–15 minutes.
2. Inspect individual stalled signals with saber signal get <signalId> to understand what's happening.
3. Export partial — proceed to Stage 5/6 with what's completed and flag the gaps to the user.

Re-running a specific stalled signal outside the subscription (for one domain + question) can be done with saber signal --domain <d> --question "<q>" --answer-type <t> --yes — this uses the domain+question cache and costs 2 credits per signal, not a full list re-run.

---

Stage 5: Contact Search

Step 1: Extract company LinkedIn URLs

saber list company companies <listId> --json

From the JSON response, each company has:

• domain: e.g. booking.com
• handle: e.g. company/booking.com → construct URL: https://www.linkedin.com/company/booking.com
• enrichedData.liId: e.g. 11348 → used later to match contacts back to companies

Build two mappings:

• handle → domain (for passing to contact search)
• liId → domain (for matching contact results back)

Step 2: Ask for contact filters

Ask the user:

• "What job titles should I look for? (e.g. CEO, VP Sales, Head of Marketing)"
• "Any country filter for contacts? (optional)"
• "Any keyword filter? (optional)"

Step 3: Run contact search

The contact search API accepts a maximum of 10 company LinkedIn URLs per call. Batch the companies into groups of 10 and run sequentially, collecting all results:

batches = [company_urls[i:i+10] for i in range(0, len(company_urls), 10)]
for batch in batches:
    # saber contact search --company-linkedin <url> ... --title CTO --json

Single-batch example:

saber contact search \
  --company-linkedin "https://www.linkedin.com/<handle1>" \
  --company-linkedin "https://www.linkedin.com/<handle2>" \
  [--title "<title>" --title "<title>"] \
  [--country "<code>"] \
  [--keyword "<keyword>"] \
  --json

Step 4: Group contacts by company (max 2 per company)

Parse the JSON response contacts[]. Each contact has a positions[] array. The first position's companyUrl is formatted as https://www.linkedin.com/company/<liId>. Extract the numeric ID and match it to the liId → domain mapping.

For each domain, keep the first 2 contacts returned. For each contact, record:

• fullName
• positions[0].title (current job title)

---

Stage 6: Combined Export

Step 1: Export companies + signals to CSV

saber list company export <listId> \
  --signal-template-id <templateId1> \
  --signal-template-id <templateId2> \
  [--signal-template-id <templateId3>] \
  --output /tmp/prospecting-workflow-signals.csv

Step 2: Merge contact columns using Python

Run this via the Bash tool, substituting the contact_map dict with real data from the contact search results:

import csv, re

# contact_map built from contact search: { "domain.com": [{"name": ..., "title": ...}, ...] }
contact_map = { ... }

with open('/tmp/prospecting-workflow-signals.csv', 'r') as f:
    rows = list(csv.DictReader(f))

if not rows:
    print("No data to export")
    exit()

fieldnames = list(rows[0].keys()) + [
    'contact1_name', 'contact1_title',
    'contact2_name', 'contact2_title'
]

import os
desktop = os.path.expanduser('~/Desktop/prospecting-workflow-export.csv')

with open(desktop, 'w', newline='') as f:
    writer = csv.DictWriter(f, fieldnames=fieldnames)
    writer.writeheader()
    for row in rows:
        # Extract domain from website column
        website = row.get('website', '')
        domain = re.sub(r'^https?://', '', website).split('/')[0]
        contacts = contact_map.get(domain, [])
        row['contact1_name']  = contacts[0]['name']  if len(contacts) > 0 else ''
        row['contact1_title'] = contacts[0]['title'] if len(contacts) > 0 else ''
        row['contact2_name']  = contacts[1]['name']  if len(contacts) > 1 else ''
        row['contact2_title'] = contacts[1]['title'] if len(contacts) > 1 else ''
        writer.writerow(row)

print(f"Saved to {desktop}")

Step 3: Confirm to the user

"Export saved to ~/Desktop/prospecting-workflow-export.csv — [N] companies, [M] signal columns, contact columns included."

---

Notes

• Never display or log API keys. The CLI authenticates automatically via stored credentials.
• Industry strings must exactly match the Saber Industries list (lowercase, case-sensitive). When uncertain, tell the user and ask them to confirm the exact string.
• Always include --yes on saber signal, subscription create, and subscription trigger to skip the interactive credit prompt.
• Export after all signals complete — partial exports leave blank cells for in-progress signals.
• Contact matching uses the numeric LinkedIn ID (liId) from the company list response matched against the companyUrl in each contact's positions. If a contact can't be matched to a company, skip it.
• Size filter format is strict: use 501-1K not 501-1000, 1K-5K not 1000-5000.