Skill instructions
company-due-diligence3.14 KB
View saved version →
---
name: company-due-diligence
description: Runs a structured KYB / due-diligence review of a company — resolves the entity, then pulls official-registry details, directors, shareholders, corporate group structure and financials into one report. Use when the user asks to run due diligence, a KYB check, a compliance review, or to "check out" / "vet" a company.
---
# Company Due Diligence
Produce a consistent KYB dossier for a single company from Global Database. Resolve
the entity once, then gather registry and financial facts and present them as a
structured report. Never infer missing values — state them as `Not Available`.
## Trigger Examples
- Run due diligence on Acme Ltd.
- Do a KYB check on acme.com.
- Vet this company before we onboard them: Acme Ltd, UK.
## Workflow
1. **Resolve the entity.**
- If the user gave a website or LinkedIn URL, use `get_company_by_website` or
`get_company_by_linkedin`.
- Otherwise use `get_company_by_identifiers` (name, registration number, VAT,
ticker, email) to obtain a `company_id`.
- If multiple matches come back, show them in a short table (name, country,
registration number) and ask the user to pick one before continuing.
2. **Official registry record.** Call `kyb_search` if you still need the registry
identifier, then `kyb_company_details` for the authoritative registry record
(legal name, status, incorporation date, registered address).
3. **People.** Call `kyb_officers` for directors, secretaries and officers. Flag any
officer whose status is not active.
4. **Ownership.** Call `kyb_shareholders` for shareholders and `kyb_group_structure`
for the corporate group. Note any ownership the registry marks as beneficial.
5. **Financials.** Call `kyb_financial` for filed statements across years. If the
registry has none, fall back to `get_company_financials`. Report the latest year
plus one prior for trend.
6. **Assemble the dossier** using the report layout below.
## Progress Display
Before each step, reprint the stepper with the entity name as a subtitle, e.g.
`> Due diligence for **Acme Ltd**`. Icons: ✅ done · ⏳ in progress · ⬜ not started
· ❌ failed.
```
⬜ Step 1: Resolving Entity
⬜ Step 2: Registry Record
⬜ Step 3: Officers
⬜ Step 4: Ownership
⬜ Step 5: Financials
⬜ Step 6: Report
```
## Report Layout
```
# Due Diligence — <Legal Name>
## Identity
- Registry status, registration number, incorporation date, registered address
## Officers
- One row per officer: name, role, appointed date, status
## Ownership
- Shareholders with % held; group parent/subsidiaries if any
## Financials
- Latest year vs prior: revenue, net profit, total assets, key ratios
## Flags
- Anything notable: inactive status, resigned officers, missing filings
```
## Rules
- Resolve the entity **once**; reuse the same `company_id` for every later call.
- Do not call tools the workflow does not list unless the user explicitly asks.
- If a step returns an error, retry up to 3 times, then record that section as
`Not Available` and continue — do not abort the whole dossier.
- Never fabricate registry numbers, ownership percentages or financial figures.
company-snapshot2.51 KB
View saved version →
---
name: company-snapshot
description: Produces a short profile of one company — identity, size, financial headline and digital footprint — without the full KYB dossier. Use when the user asks "what is this company", "tell me about", "look up" or "profile" a company, or pastes a company website or LinkedIn URL and asks who they are.
---
# Company Snapshot
A fast read on a single company. This is the light path: identity, headline
financials and digital footprint, in one screen. When the user asks for due
diligence, a KYB check or anything compliance-shaped, use
`company-due-diligence` instead — it pulls the official registry record.
Never infer a missing value. Report it as `Not Available`.
## Trigger Examples
- What is acme.com?
- Tell me about Acme Ltd.
- Look up this company: linkedin.com/company/acme
- Profile Acme before the call.
## Workflow
1. **Resolve the entity, once.**
- Website → `get_company_by_website`.
- LinkedIn URL → `get_company_by_linkedin`.
- Anything else (name, registration number, VAT, ticker, email) →
`get_company_by_identifiers`.
- If several companies come back, show name, country and registration number
in a short table and ask which one before continuing.
- Keep the resulting `company_id` and reuse it for every later call. Do not
re-resolve.
2. **Profile.** `get_company_details` for the firmographic record.
3. **Financials.** `get_company_financials`. Report **the latest year and one
prior** — nothing more. The payload carries every year on file and dumping it
all buries the answer.
4. **Digital footprint.** `get_digital_insights` for traffic, ranking and
technology stack. Skip this step if the user only asked "who are they".
## Report Layout
```
# <Legal Name>
<one-sentence description of what the company does>
| | |
|---|---|
| Country | |
| Founded | |
| Employees | |
| Industry | |
| Website | |
## Financials
Latest year vs prior: revenue, net profit, total assets.
## Digital
Traffic rank, main technologies.
## Sources
```
## Rules
- Resolve **once**; reuse the same `company_id`.
- Do not call `kyb_*` tools here. Those belong to `company-due-diligence`, cost
a registry lookup, and are not what the user asked for.
- Never compute a column the payload does not contain. If revenue per employee
is wanted, say it is derived and show the two inputs.
- Cite the source for every figure that carries one.
- If a step errors, retry up to 3 times, then mark that section `Not Available`
and continue. Do not abandon the snapshot.
prospect-list3.12 KB
View saved version →
---
name: prospect-list
description: Builds a target account list from firmographic filters — country, industry, size, revenue — then optionally finds and enriches contacts at each company. Use when the user asks to find, search or build a list of companies matching criteria, to source leads or accounts, or to get contacts at a set of companies.
---
# Prospect List
Turn a description of an ideal customer into a concrete list of companies, and
optionally into named contacts. One company is a job for `company-snapshot`;
this skill is for sets.
## Trigger Examples
- Find SaaS companies in Germany with 50–200 employees and over €10M revenue.
- Build me a list of UK fintechs founded after 2020.
- Who should we target in the Nordics for logistics software?
- Get me contacts at these 20 accounts.
## Workflow
1. **Settle the filters before searching.** Restate the criteria back to the
user as a short list and confirm anything they left open — country, industry,
employee band, revenue band, founding year. A vague search burns a call and
returns noise.
2. **Resolve filter values.** Call `get_nomenclature` to get the accepted values
for industry, country and any coded field. Do **not** guess a code or pass a
free-text industry name — filters are matched against the nomenclature, and an
invented value silently returns nothing rather than erroring.
3. **Search.** Call `prospecting` with the resolved filters. Report how many
companies matched **before** listing any. If the count is very large, say so
and offer to tighten the filters rather than paging through it.
4. **Present the list.** One row per company: name, country, industry,
employees, revenue, website. Stop at 25 rows unless the user asked for more,
and say how many were withheld.
5. **Contacts — only if asked.** Do not go looking for people unless the user
wants them.
- For a company already resolved: `search_employees`, filtered by role or
seniority when the user named one.
- Starting from a domain instead: `find_people_by_domain`.
- For a full profile of one person: `get_employee_details`.
- `enrich_employee_contacts` returns email and phone. Call it **per contact
the user actually wants**, not across the whole list — it is the expensive
step and most rows will never be worked.
## Report Layout
```
# Target list — <criteria in one line>
<N> companies matched. Showing <M>.
| Company | Country | Industry | Employees | Revenue | Website |
|---|---|---|---|---|---|
## Contacts (if requested)
| Company | Name | Role | Email | Phone |
```
## Rules
- Never invent a filter value. Resolve it through `get_nomenclature` first.
- Never fabricate an email address or phone number. If enrichment returns
nothing, the cell is `Not Available` — a plausible-looking guess at a work
email is worse than a blank.
- State the match count before the rows, and state what was truncated.
- Enrich contacts one at a time, on request. Do not bulk-enrich a list.
- If the search returns zero, say which filter is most likely responsible and
propose a specific loosening — do not silently retry with different criteria.