Import Prospects

Import Prospects - Bring Your Own List

A skill that imports existing prospect / contact lists (CSV, Excel, SQLite, plain text, etc.) into LeadAce. The MCP server only accepts a canonical CSV (specific columns), so the LLM here is responsible for reading the user's arbitrary file, mapping it to the canonical schema, writing the canonical CSV, and uploading it.

Before starting: Read ${CLAUDE_PLUGIN_ROOT}/references/workspace-conventions.md and follow the cross-cutting conventions there (data storage, MCP error handling, document writes, output discipline).

Two import modes

• Tenant assets only (default for "name card lists" / "old CRM exports" / lists not tied to a single campaign): no projectId is sent to the upload tool. Prospects + organizations are saved at the tenant level. Use /match-prospects <project> afterwards to link the right ones into a specific project.
• Linked to a project: a projectId is sent; every row also carries a matchReason and is linked to that project via project_prospects. Use this when the list was assembled specifically for one campaign.

Inputs

• $0 — path to the source file (required). Any tabular format works (CSV, TSV, XLSX, XLS, ODS, SQLite, plain text with consistent delimiters).
• $1 — project name (optional). If supplied, default to project-linked mode (still confirm with the user in step 1). Must already exist; create one with /leadace <your-homepage-URL> first.
• $2 — dedup policy: skip (default) or overwrite. With skip existing rows are left alone; with overwrite matched rows have their fields refreshed (and, in project mode, re-linked to the project).

Canonical CSV schema

* At least one of email, contactFormUrl, any snsAccounts.*, or platformUrl is required per row. ** Tenant-only imports may omit the matchReason column entirely. Linking happens later via /match-prospects, which writes a fresh matchReason per project.

Maximum 1000 data rows per import. Split larger files into multiple calls.

Steps

1. Decide the import mode

Use AskUserQuestion to pick between the two modes — do not assume. Default the first option to:

• "Save as tenant assets only" when $1 is empty.
• "Link to a project" (with the named project pre-selected) when $1 is provided.

Options:

• Save as tenant assets only — prospects are stored at the tenant level. Use this for name-card / old-CRM / general lead lists where the right project is unclear. Linking happens later with /match-prospects.
• Link to a project — every row is linked to a project via project_prospects. Use this when the list was assembled for one specific campaign.

If the user picks Link to a project:

• if $1 is given, use it as PROJECT_NAME.
• otherwise call list_projects and ask via AskUserQuestion (or use the only project if there's just one).
• if there are zero projects, abort: "No projects yet. Run /leadace <your-homepage-URL> first, or re-run this skill in tenant-asset mode."

If the user picks Save as tenant assets only, set PROJECT_NAME = null and skip every project-resolution step below.

2. Read the source file

If $0 is empty, abort and ask the user for a file path.

Inspect the file by extension:

• .csv / .tsv / .txt — Read directly
• .xlsx / .xls / .ods — convert to CSV via Bash using whatever the user's environment has (python3 -c "import pandas; pandas.read_excel('$0').to_csv('/tmp/leadace_import_src.csv', index=False)" is a reasonable default; ssconvert from gnumeric works too). If conversion is impossible, tell the user the install hint and abort.
• .sqlite / .db — sqlite3 $0 .schema to see tables, ask the user which table to use (AskUserQuestion), then sqlite3 -header -csv $0 'SELECT * FROM <table>;'
• Anything else — try Read and let the LLM infer the structure

Read enough rows to understand the schema (first ~50 rows is plenty). Do not attempt to read multi-MB files in one shot.

3. Map columns to the canonical schema

For each source column, decide what canonical column it corresponds to. Do not invent data:

• If organizationDomain is missing but you have an email or website URL, derive the apex domain (e.g. https://blog.acme.com/foo → acme.com).
• If organizationWebsiteUrl is missing, derive https:// + organizationDomain.
• If websiteUrl is missing, fall back to organizationWebsiteUrl.
• If overview is missing, synthesise a one-line description from available fields (industry, name, role) — never leave it blank.
• matchReason handling depends on the mode chosen in step 1:
  • Tenant-only mode: omit the matchReason column entirely from the canonical CSV. Don't ask the user — /match-prospects will write fresh per-project reasons later.
  • Project-linked mode: if the source has no usable matchReason, ask the user once for a default ("Why is this list a fit for this project?") and apply it to every row.
• country (recommended): map it when the source carries it, with countrySource manual (source/user-given) or ai_inferred (you derived it). If the source has no country at all, don't leave it blank silently — fetch list_country_codes (recognized codes + send-allowed flags) and ask the user once whether to apply one default country to all rows or leave them blank. Present that fetched list; don't invent one.
• industry: source labels rarely match the controlled vocabulary verbatim — fetch tpl_industries via get_master_document and map each source value to its closest entry (Other when none fits). Leave the cell blank rather than passing a free-form label through; unmapped values reject the row.
• doNotContact mapping (important): if the source has any do-not-contact-like column — common names: do_not_contact, dnc, opted_out, opt_out, unsubscribed, unsubscribe, is_unsubscribed, subscribed=false — map it to canonical doNotContact. Emit 1 for truthy values and leave the cell blank for falsy ones. Do not silently drop these rows. They must be imported with the flag set; otherwise /build-list may rediscover the same organisation later and the user will outreach to someone who already opted out. Truthy: 1, true, yes, on, unsubscribed, opted out. Falsy: 0, false, no, off, subscribed, active, blank. If the semantics are ambiguous, ask the user once before mapping.
• If a row has none of email / contactFormUrl / snsAccounts.* — drop it and report it in the per-row error summary.

If the source has columns you cannot place anywhere, ignore them — do not try to widen the schema.

4. Confirm with the user before uploading

Show the user:

• the import mode chosen (tenant-only vs linked to project <name>)
• the source path and row count detected
• the column mapping you inferred (source column → canonical column)
• if a doNotContact column was mapped, how many rows will be flagged DNC vs not
• any rows you are dropping and why

Ask via AskUserQuestion whether to proceed. If the user wants tweaks, iterate.

5. Write the canonical CSV

Write the canonical CSV to /tmp/leadace_import_<scope>_<timestamp>.csv using Write (use the project name as <scope> for project-linked mode, or tenant for tenant-only mode). Use proper CSV quoting:

• Fields containing ,, ", or newlines must be wrapped in "...".
• A literal " inside a quoted field is doubled ("").

Header row must be exactly the canonical column names (case-sensitive).

6. Pick the dedup policy

If $2 is skip or overwrite, use it directly. Otherwise ask via AskUserQuestion:

• skip (default, safer) — existing prospects untouched
• overwrite — matched prospects get their fields refreshed (and, in project-linked mode, re-linked to that project)

7. Upload

Read the canonical CSV file back into a string and call import_prospects_from_csv:

• Project-linked mode: pass projectId: PROJECT_NAME plus csvText and dedupPolicy.
• Tenant-only mode: omit projectId entirely. Pass only csvText and dedupPolicy. Do not pass an empty string — leave the field out.

8. Report

Surface the counts the tool returns:

• inserted — newly created
• overwritten — existing prospects updated (only with overwrite)
• skipped — duplicates / do_not_contact / plan limit
• errors — rows that failed validation

For non-zero skipped or errors, summarise the top reasons. If the free plan limit was hit, advise the user to upgrade or delete unused projects.

In tenant-only mode, finish by suggesting the next step: Run \/match-prospects ` to surface the imported prospects that fit a specific project's strategy and link them.`

Leave the canonical CSV at /tmp/leadace_import_*.csv so the user can re-run with a different policy if needed.