--- name: gtm-prospecting-desk description: "This skill should be used when the user wants to run the Outbound Pipeline Engine skill, or asks to \"build my prospecting pipeline\", \"run my outbound research\", \"keep a prospect table current\", or wants recurring ICP-filtered company and people discovery with scored rows and outreach drafts. Runs on the superagnt MCP: first-party people/company data + LinkedIn + workspace database. Drafts only: sending is always the human's." version: 0.2.0 --- # 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.