Outbound Pipeline Engine

Outbound research end to end, compounding in one pipeline table. This skill builds on the base lead-generation skill's workflow — read that one for the per-step tool detail; this one adds the standing-desk structure.

Step 0 — tools

agnt_tools_list_enabled; if missing: agnt_tools_enable({ "families": ["data:linkedin", "database"] }) (the first-party data_agnt_* people/company tools are always on). requires_upgrade → confirm_url to the user → wait → re-run. (The gtm-prospecting-desk skill endpoint pre-enables these.)

Tool names on grouped servers

This skill writes flat (per-operation) tool names. New superagnt servers list grouped tools instead: one tool per resource with an action argument. Flat names still resolve there but are not listed, so call the grouped form:

• data_agnt_companies_search / data_agnt_companies_discover → data_agnt_companies, action search / discover
• data_agnt_people_search / data_agnt_people_enrich → data_agnt_people, action search / enrich
• data_agnt_people_email_finder / data_agnt_people_email_verifier → data_agnt_contacts, action find_email / verify_email
• data_linkedin_* profile reads → data_linkedin_profiles (e.g. action profile_by_url)
• agnt_db_insert / agnt_db_upsert → agnt_db_write, action insert / upsert
• agnt_db_select → agnt_db_read, action select
• agnt_agents_create → agnt_agents_write, action create
• agnt_schedules_create → agnt_schedules_write, action create
• agnt_tools_enable → agnt_tools_write, action enable

On a grouped server every data source is already on, so skip the data:* part of Step 0 (enable any other family with agnt_tools_write, action enable). Grouped data calls return a compact markdown view: pass response_format: "json" for every view field, or "raw" for the full payload, when a step needs a field the table leaves out.

Step 1 — pin the ICP (first run only)

Company filters (size, industry, geography, signals) + buyer titles + any disqualifiers, into icp_profile (agnt_db_insert). Create prospects(id, company, company_domain, person, title, linkedin_url, email, email_verified boolean, fit_score int, fit_reason text, status text, draft text, updated_at timestamptz) — upsert key (person, company_domain).

Step 2 — the desk run

1. data_agnt_companies_search against the ICP filters (or data_agnt_companies_discover for similar-to expansion); skip companies already in the table unless refreshing.
2. data_agnt_people_search per company for the buyer titles; data_agnt_people_enrich on the chosen person; data_linkedin_* for profile detail when the LinkedIn family is enabled.
3. data_agnt_people_email_finder then ALWAYS data_agnt_people_email_verifier. Unverified rows keep email_verified=false — they stay workable via LinkedIn.
4. agnt_db_upsert every row with status transitions (new → enriched → scored → drafted) so a failed run resumes instead of restarting.
5. Score each row (fit_score + one-line fit_reason from the ICP), then write outreach drafts into draft for the top rows — the user's voice (ask for two past emails once, store the voice notes in icp_profile).

Step 3 — report

From the table: counts by status, the top 10 by fit with reasons, and where the drafts are. Numbers come from agnt_db_select, never memory.

Step 4 — offer the standing desk (a task agent)

After one successful run: offer a scheduled refresh (weekly beats daily for prospecting) by building a task agent — recipe in the base task-agents skill. agnt_agents_create with a system prompt carrying step 2 verbatim (ICP table, pipeline table, status transitions), only the tools it uses, deploy, then agnt_schedules_create; cadence confirmed with the user, billed runs stated plainly, and a plan limit's requires_upgrade handed to the user as its confirm_url. 500+ row enrichment belongs in a data job (see data-pipelines) rather than a longer session.

Hard rules

• Drafts only. This skill never sends email, LinkedIn messages, or anything else. Wiring a sending tool is the user's own deliberate act.
• Respect disqualifiers in icp_profile absolutely — a disqualified company never re-enters the pipeline.