moonlight-lupin/agent-skills / research/entity-research
entity-research
VerifiedUse when the user says "research this company / person", "background check on X", "any negative press / adverse media on X", "who owns / who runs X", "is X sanctioned / sanctions check on X", "vet this vendor / counterparty / candidate / partner", or "entity search" — deep background research on an entity (a company OR a person) into a cited dossier: identity & background, ownership & key management, adverse / negative media, public sanctions-list name-match signals, PEP indications from public research, and litigation / regulatory history. It RESEARCHES and COMPILES with sources; it is NOT a compliance/AML/CDD determination and NOT a sanctions or PEP clearance — a sanctions-list or PEP signal is a SIGNAL to escalate to a human compliance function, never a "clear" or "block". Research a person only for a legitimate purpose and only from public information.
| File | Kind | Size |
|---|---|---|
| README.md | markdown | 1.8 KB |
| SKILL.md | skill md | 13.0 KB |
| references/boundaries-and-sanctions.md | markdown | 4.5 KB |
| references/organisation-evidence-exa.md | markdown | 5.7 KB |
| references/research-lenses.md | markdown | 3.7 KB |
| scripts/entity_research.py | script | 17.1 KB |
| tests/__init__.py | script | 0 B |
| tests/test_entity_research.py | script | 23.4 KB |
references/boundaries-and-sanctions.md
View on GitHub# Boundaries & sanctions — signal, not determination
The single most important rule for this skill: it **researches and compiles**; it does **not
determine, screen, clear, rate or block**. Get this wrong and it impersonates a regulated function (a
CDD/AML screening function, a compliance decision).
## The signal-not-determination rule
- A sanctions / PEP / watchlist name-match is a **potential-match SIGNAL** → **escalate to your
compliance / AML function** for verification. Never write "X is sanctioned" or "X is clear".
- **No match is NOT a clearance.** The lists aren't exhaustive and the matcher is basic (below).
Absence of a hit means *this check found nothing* — nothing more.
- Negative media is **allegation + source + date**, not a finding. Distinguish allegation → charge →
outcome. Avoid defamatory framing; report what the source says, attributed.
## The lists (`screen_lists`)
Official government consolidated lists, fetched live and cached (`~7 days`):
- **OFAC-SDN** — US Treasury Specially Designated Nationals (≈19k).
- **OFAC-CONS** — US Treasury Consolidated (non-SDN) list (≈0.4k).
- **UK-OFSI** — UK Sanctions List, published by FCDO/OFSI (≈59k rows incl. aliases; *this is what
catches a UK-only designation an OFAC/UN check misses*). The old OFSI Consolidated List feed
(ConList.csv) closed 2026-01 and now returns 409; this is issue #31's resolution.
- **UN** — UN Security Council consolidated list (≈1k).
**National / local lists.** Many countries maintain their own autonomous lists, and several have **no
clean machine-readable feed**. For the subject's home jurisdiction, do a **manual portal check** of
the relevant authority's site — and don't represent the absence of a feed as having fully screened
that jurisdiction.
*Extensible:* EU and others — add a fetcher/parser + URL to `LIST_URLS` / `_PARSERS` and
`DEFAULT_LISTS`. (EU's official download needs a free token, so it isn't bundled.)
URLs can drift (OFAC has migrated endpoints before) — if a list shows as `unavailable`, update the URL
constant or do an official-portal web check for it. Fetch failures degrade gracefully (the list is
skipped and reported in `unavailable`, with a fallback to stale cache).
## Matcher limits (why it's only a signal)
`screen_lists` does **token-based** matching (normalised, corporate-suffix-stripped):
- **Catches:** exact and token-subset matches ("Northwind Bank" → "NORTHWIND BANK"; "Acme Trading" →
"ACME TRADING CO") and strong multi-token overlap.
- **Misses:** **transliteration / spelling variants** (e.g. a name romanised differently),
**phonetic** matches, heavy typos, and single-token names listed under a variant spelling. It is
**not** fuzzy/phonetic like a professional tool (World-Check, Dow Jones).
→ This is precisely why a no-match is not a clearance, and why a **professional screening tool /
your compliance function is authoritative**. Use this for a fast first signal only.
## False positives
Common names produce matches that aren't your subject. Before flagging, sanity-check the matched entry
against your subject's identifiers (jurisdiction, type, DOB, program). Present a match as "potential
match on '<matched name>' (list, score) — to verify", with the detail, so a reviewer can adjudicate.
**Generic-word guard.** A match needs at least one **distinctive** shared token — generic business
words (`capital`, `management`, `partners`, `investment`, `group`, `global`, `trading`, `properties`,
…; see `_GENERIC`) don't create a match on their own. This stops "X Capital Management" from matching
an unrelated "Y Capital Management". Screen on the **distinctive** part of a name where possible (e.g.
"Northwind", not "Northwind Capital Management").
## Persons — privacy guardrails
- Research a person **only for a legitimate purpose** (vetting a counterparty, hire, partner) and only
**publicly-available** information.
- Don't compile **sensitive** personal data (health, religion, sexuality, political views) or build an
intrusive profile; stay on integrity/role/affiliation relevant to the purpose.
- Keep the dossier **local**; it's internal and may name individuals.
## Data handling — search the name, not the relationship
A bare name with no client/deal link is **not sensitive** to search. But never put your **relationship**
into an external query ("our client X", "we're buying X"). If the subject is tied to a live deal or a
client, that *relationship* stays confidential (omit it from queries); the public research on the name
still proceeds.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| add-product-e2e-evalpaperclipai/paperclipAdd or extend a Paperclip full-stack runner E2E workflow, fixture, matcher, or report evidence path for local or Daytona execution. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |
| add-runner-evalpaperclipai/paperclipAdd or extend a Paperclip Runner protocol evaluation definition, roster, assertion, or report fixture with provenance and narrow validation. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |
| diagnose-why-work-stoppedpaperclipai/paperclipCompatibility entry point for failed, stalled, looping, or over-recovered Paperclip work. Use fix-failed-paperclip-run for the full diagnosis, regression, repair, real verification, and green-PR workflow; honor an explicitly requested diagnosis-only scope. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |
| fix-failed-paperclip-runpaperclipai/paperclipDiagnose and fix failed, stuck, slow, or incorrectly blocked Paperclip tasks and runs. Use with a task/run URL or failure report to trace the root cause, prove a regression red then green, check second-order effects, verify the real workflow, and deliver a PR with passing checks. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |