Competitor Tracker

Measure posts-per-hour and engagement for a fixed brand set with an identical method every day, so the trend is real. The methodology is the one from the public superagnt research reports: identical queries, literal-match filtering, and rates computed from the returned page's own timestamps.

Step 0 — tools

agnt_tools_list_enabled; if missing: agnt_tools_enable({ "families": ["data:x", "database"] }). requires_upgrade → hand the confirm_url to the user, wait, re-run. (The competitor-mindshare skill endpoint pre-enables these. The workspace dashboard's tools, agnt_dashboards_*, are on for every workspace.)

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_write, action insert
• agnt_db_apply_migration → agnt_db_sql, action apply_migration
• agnt_dashboards_guide / agnt_dashboards_preview → agnt_dashboards_read, action guide / preview
• agnt_dashboards_write_file / agnt_dashboards_publish → agnt_dashboards_write, action write_file / publish
• 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 — name the brands (first run only)

The user lists brands/products to track (include their own). Record query variants per brand — people abbreviate product names — in mindshare_brands(id, brand, variants text[], active). Create mindshare_daily(id, brand_id, measured_at timestamptz, posts_per_hour numeric, median_likes numeric, sample_size int, top_post_url text) via agnt_db_apply_migration.

Step 2 — measure

Per brand, identically: data_x_search_get_search_search (section=latest, fixed like-floor, same limit for every brand). Filter the returned page CLIENT-SIDE to posts whose text literally contains a brand variant — the endpoint fuzzy-matches, and skipping this step inflates everything. Rate = matching posts ÷ the span between the page's first and last timestamps. Median likes over the matches. One agnt_db_insert row per brand per day.

Method caveats to state in any summary: mindshare ≠ market share; gaps under ~1.5x are noise; generic brand names (a common word) over-count and need tighter variants.

Step 3 — the scoreboard view

First run: agnt_dashboards_guide once (the files, the query grammar, the sa-* elements and publishing). Then build one view, mindshare, in your draft with agnt_dashboards_write_file (pass intent on the first write):

• queries/mindshare_today.sql: today's posts per hour and median likes per brand, highest first.
• queries/mindshare_trend.sql: one row per day, oldest first, one column per brand (pivot in SQL: max(d.posts_per_hour) filter (where b.brand = 'Acme') as acme).
• views/mindshare.html: an sa-chart kind hbar over today's rates, an sa-chart kind line over the trend (y-fields the brand columns, series-labels the brand names), and an sa-table of the latest measurements.
• the view's entry in dashboard.json.

agnt_dashboards_preview the view, fix what its report names, then agnt_dashboards_publish. Later runs just insert rows; the view stays current on its own (a new brand needs its column in the trend query, then publish). To share the scoreboard, the user turns on the view's public link from the superagnt app: agents can't make one.

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

Same pattern as every packaged skill: once one measurement pass has succeeded, offer a daily run by building a task agent (recipe: the base task-agents skill) — agnt_agents_create with a system prompt carrying step 2 verbatim (brand table, query shape, insert), only the tools it uses, deploy, then agnt_schedules_create (cadence confirmed; billed runs; a plan limit comes back as requires_upgrade with a confirm_url for the user). Local-only users just re-run the skill.

Rules

• Never compare numbers measured with different query shapes, floors, or limits — change the method, start a new series (add a method_version column rather than mixing).
• Every published number traces to rows in mindshare_daily.