/linkedin-growth-import

description: Import leads from a LinkedIn / Sales Navigator search into the pipeline — search, qualify against the ICP, dedupe, and store with round-robin assignment.
argument-hint: "[search URL or filters] [list name]"

Use the **linkedin-growth** skill to run a Phase A import. Arguments: $ARGUMENTS

Drive the full import flow:

1. Determine the inputs. If I gave a search URL, detect the type (URL containing `/sales/` → `nv`, otherwise `st`). If I gave filters instead, use them. Ask me for anything missing: which account should run the search (`--searcher`), and a list name (`--list`). If I have more than one account, confirm the searcher with me. 2. **Always ask me the limit** — how many of the found leads to take. I can give a number or say "max" for the maximum (Sales Navigator caps at 2500, standard search at 1000). Do not assume a default; this question is required every time. 4. Run `node scripts/import.mjs prepare --searcher <acct> --list "<name>" --type <nv|st> --limit <N|max>` with either `--search-url "<url>"` or the filter flags. (`--limit` is required; a number above the cap is clamped and reported as `limit_capped_to_max`.) 5. **Settle the ICP before qualifying** (this is the filtering step — make it explicit, not silent). The `prepare` output tells you whether an ICP is configured:

• If none is configured, interview me: which roles/seniority to target, which

industries/company types fit, company size/stage if relevant, locations to include or exclude, and any hard exclusions. Summarize what you heard, then save it with `node scripts/settings.mjs set icp_definition --stdin` (pipe the text via a heredoc — it is stored in the database, not a file; do not leave stray ICP files in the repo or tmp).

• If one is configured, show it to me in plain language and ask whether to use it as-is or

adjust it for this list (re-save if I change it). 6. Read the returned candidate file and qualify every candidate **against my ICP**, using the output contract in `config/qualification-prompt.md`. For more than ~25 candidates, chunk the work and delegate each chunk to a sub-agent. **Use a cheap/fast model for these qualification sub-agents** — it's simple classification, not deep reasoning (in Claude Code, spawn them with `model: "haiku"`; in other hosts use their small/fast model). Keep orchestration on the main model. Produce a JSON array `[{hashed_url, suitable, reasoning}]` covering every candidate (reasoning must cite the ICP criterion that drove the decision), and write it to the expected result file path. 7. Run `node scripts/import.mjs commit --batch <id> --results <file>`. 8. Report transparently: how many were found, skipped as duplicates, kept vs filtered, a few concrete sample reasons from both sides, and how the new leads were distributed across accounts.

Never send invites here — import only stores leads as `not_connected`.