Newsletter Autopilot

Research the user's topics on X and the web all week, keep only what is genuinely new, and on send day assemble a COMPLETE issue draft in their voice: sections, links, every claim cited, subject line options. The user edits and sends; sending is never the agent's.

Step 0 — tools

agnt_tools_list_enabled; if data:x, data:web or database families are missing, agnt_tools_enable({ "families": ["data:x", "data:web", "database"] }). A requires_upgrade result carries a confirm_url — give it to the user and wait, then re-run. (Connecting through the newsletter-autopilot 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_x_search_get_search_search → data_x_discover, action search
• data_web_search / data_web_scrape keep their names on both surfaces
• agnt_db_insert → agnt_db_write, action insert
• agnt_db_apply_migration → agnt_db_sql, action apply_migration
• agnt_agents_create / agnt_agents_deploy → agnt_agents_write, action create / deploy
• 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 — learn the letter (first run only)

Ask the user for the topics they cover, the send day, and ONE past issue (a link data_web_scrape can read, or pasted). From that issue record the voice and the section shape: how it opens, how many sections, how takes are framed, how links are dropped. Derive per topic 1-3 X searches with engagement floors plus 1-2 web queries. Store it:

• issue_profile(id, topics text[], send_day text, voice_notes text, section_shape jsonb, updated_at timestamptz) — via agnt_db_apply_migration then agnt_db_insert.
• issue_queries(id, topic, kind text check (kind in ('x','web')), query, min_likes int, active, created_at).
• issue_material(source_url primary key, topic, title, summary_md text, found_at timestamptz, used_in_issue text) — the week's pile + the dedupe ledger in one.
• issue_drafts(week primary key, draft_md text, subject_options jsonb, status text default 'draft', assembled_at timestamptz).

Step 2 — research all week

Each run (nightly when scheduled): X queries through data_x_search_get_search_search (section=top, the query's like floor), web queries through data_web_search. Drop anything already in issue_material. For the keepers that are pages, data_web_scrape; for standout X posts, pull the thread. The bar: a find earns its row when it would earn a sentence in the issue — new information, a real number, a take worth reacting to. Write a 2-4 sentence summary_md per keeper so assembly never re-reads sources.

Step 3 — assemble the issue

On send day (or on demand), build the COMPLETE draft from the week's unused issue_material, in the recorded voice and section shape: the open, each section written out with its links inline, every claim citing its source URL, and 2-3 subject line options. Save to issue_drafts, mark the material rows used_in_issue, and hand the draft over. The user edits and sends from their own ESP — this skill never sends, schedules, or publishes anything.

Step 4 — offer the nightly run (a task agent)

Once one issue has shipped from a draft, offer to make the research nightly by building a task agent — the full recipe is the base task-agents skill. Short form: agnt_agents_create with a system prompt carrying step 2 verbatim (tables, queries, the quality bar) and step 3 gated on the send day, add ONLY the tools those steps use, agnt_agents_deploy, then agnt_schedules_create — cadence confirmed with the user first; every firing is a billed run. The lifecycle and schedules families are on every plan, and plans cap agents and schedules; requires_upgrade → confirm_url → wait. If the user prefers local-only, skip this and tell them to run the skill through the week.

Rules

• Every claim in the draft cites its source URL — no cite, no claim.
• Quote sparingly: at most one short quotation per source; the draft summarizes in the user's voice, it never republishes someone else's writing.
• This skill drafts; it never sends email, posts, or touches the user's ESP.