القائمة
Resolve company names, company descriptions, investor/funding filters, or vertical queries into canonical TurboPuffer company IDs for downstream people search.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
The single post-import people-processing workflow and per-person dossier surface. Use for $deep-context, "process/resolve/enrich my contacts", "build deep context", a dossier or identity lookup by name/phone/email, duplicate-person review, LinkedIn self-heal, or the staged people/LinkedIn UI. Builds dossiers for imported people and unresolved Gmail/iMessage/WhatsApp candidates, merges duplicates, asks the user only about uncertain additions, runs one budget-gated lookup for the editable Yes decisions plus eligible wrong-link recovery, verifies found LinkedIns, then realizes the approved network and index.
Add Gmail contacts to your local network. Use for $import-gmail. Sets up msgvault/Gmail (OAuth + authorize), asks which accounts and how many years to sync, syncs mail down, and imports contacts free and locally (shared identity directory only) — unresolved contacts go to a research-candidates pool. No Parallel.ai, no RapidAPI, no index build — identity resolution and indexing happen later in $deep-context. Always reruns the full checklist; overwrites in place.
Add iMessage/WhatsApp contacts to your local network. Use for $import-messages. Sets up access (Full Disk Access + WhatsApp QR), syncs message contacts down, matches them against already-imported LinkedIn/Gmail people (free), and imports matched people plus a research-candidates pool. No LLM calls, no paid research, no index build — identity resolution and indexing happen later in $deep-context. Never uploads to a Powerset set.
Deterministic LinkedIn-only Powerpacks setup. Use for $setup. Steps through update, a supported Powerset workspace or an already-prepared custom Modal workspace, LinkedIn Connections.csv import on Modal, source merge, Modal indexing, and local search-index validation. Always reruns the full checklist; overwrites in place. For Gmail use $import-gmail; for iMessage/WhatsApp use $import-messages.
Bootstrap Powerpacks into this agent from one URL — clone the public repo, install every Powerpacks skill for this harness (Claude Code, Codex, or Pi), initialize the hosted Powerset config when requested, then continue the user's ask in the same session without a restart. Use for "install powerpacks", "download the powerpacks skill", "use powerpacks to set up ...", or "set up my local network search using my Powerset account".
Force-update the canonical Powerpacks checkout to the newest published release and reinstall skills while preserving .powerpacks. Use for $update-powerpacks.
| name | search-company |
| description | Resolve company names, company descriptions, investor/funding filters, or vertical queries into canonical TurboPuffer company IDs for downstream people search. |
Use this when the user asks for company search directly, or when they want the company set behind a people search inspected first.
Examples:
/search-company database companies/search-company series b fintech startups in california/search-company companies backed by sequoiaFresh company resolution only: do not scan, discover, reuse, or resume prior company search artifacts, state files, CSVs, JSONL, or manifests. Resolve the company set from the user's current request every time.
role-search-filters company fields only:
company_names, company_semantic_queries, entity_types, sector_types,
funding/headcount filters, company geography, investor_names, and
resolved investors. Preserve set_id when the user provides one.set_id, run resolve_set_operators without --set-id so it inherits
POWERPACKS_DEFAULT_SET_ID / POWERSET_DEFAULT_SET_ID, or falls back to the
logged-in operator's active personal set. Copy the returned operator_ids
into the company resolver payload. Do not pass the raw set UUID as
allowed_operator_ids.investor_names are present, run resolve_investors first and record or
copy the returned investor_urns into investors.resolve_companies.company_ids and
operator_ids into
search; for large company sets, use apply_prefilters so people
search batches company IDs instead of sending one giant company_id In [...].Use company_semantic_queries for vertical/domain intent, not just
sector_types. Coarse sectors are useful, but the resolver needs an explicit
sector strategy:
company_sector_strategy: "staged" by default for ambiguous domain
searches. This starts with sector-intersected semantic search and broadens if
the hard frontier is too small.company_sector_strategy: "soft_union" when recall matters more than
precision or labels are known to be incomplete. This matches semantic results
OR sector-filter matches.company_sector_strategy: "hard_filter" when the user explicitly wants
sector labels as a strict constraint.company_sector_strategy: "semantic_only" when sector labels are too
noisy for the requested company domain.Good:
{
"company_semantic_queries": [
"Companies building database systems, hosted databases, data storage engines, database infrastructure, SQL or NoSQL databases, or developer platforms for managing application data."
],
"sector_types": ["data"],
"company_sector_strategy": "staged",
"company_sector_min_results": 500
}
Bad:
{ "sector_types": ["data"] }
uv run --project . python packs/search/primitives/resolve_set_operators/resolve_set_operators.py \
--payload-json '<json>' \
--env-file .env
uv run --project . python packs/search/primitives/resolve_investors/resolve_investors.py \
--payload-json '<json>' \
--env-file .env
uv run --project . python packs/search/primitives/turbopuffer/turbopuffer_resolve_companies.py \
--payload-json '<json>' \
--env-file .env