Content Ideas Engine

Mine X for what the user's audience actually talks about; file the signal; write a cited ideas report. The recipe compounds: every sweep builds on the tables the last one left.

Step 0 — tools

agnt_tools_list_enabled; if data:x or database families are missing, agnt_tools_enable({ "families": ["data:x", "database"] }). A requires_upgrade result carries a confirm_url — give it to the user and wait, then re-run. (Connecting through the audience-radar 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
• agnt_db_insert / agnt_db_upsert → agnt_db_write, action insert / upsert
• agnt_db_select → agnt_db_read, action select
• 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 — define the radar (first run only)

Ask the user for their niche, audience, and 3–5 accounts they rate. Derive 8–12 standing searches: topic phrases, "how do I"-shaped questions, tool names, and the rated accounts' handles. Sensible engagement floors (10–50 likes; higher for noisy topics). Store them:

• radar_queries(id, query, min_likes, active, created_at) — via agnt_db_apply_migration then agnt_db_insert.
• radar_posts(post_id primary key, query_id, author, text, likes, replies, url, posted_at timestamptz, seen_at timestamptz).

Step 2 — sweep

For each active query: data_x_search_get_search_search (section=top, the query's like floor). Upsert results into radar_posts on post_id (agnt_db_upsert) so re-runs never duplicate. Pull reply threads only for standout posts — replies are where unmet needs hide, and thread pulls cost more credits than searches.

Step 3 — report

From radar_posts since the last sweep (agnt_db_select): the patterns that repeated, 3–5 posts worth replying to today, and 3–5 post ideas — EVERY idea cites the post that earned its place (author + link + likes). No cite, no idea. Keep the report under a page.

Step 4 — offer the night shift (a task agent)

Once one sweep has succeeded end to end, offer to make it 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 steps 2–3 verbatim (tables, queries, report shape), add ONLY the tools those steps use (agnt_tools_search for the config refs), 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 re-run the skill whenever.

Rules

• Never fabricate engagement numbers or post text — everything reported comes from a tool result or the table.
• This skill reads X; it never posts, replies, or DMs.