This is a Claude Code plugin by Sabahudin Murtic. Install it from his repo — https://github.com/sabahudin-web/competitive-intelligence-radar — which bundles the commands, agents, skills, and rules described here. It needs a BrightData API token (live web data) and a Notion connection (publishing), and the optional war room uses Claude Code's experimental agent-teams feature.

The three commands

/radar-setup — one-time onboarding. Asks whether BrightData and Notion are connected and live-tests one real call to each (it never assumes), collects your business context (domain, tracked keywords, competitor list, geo), shows the proposed Notion schema and creates the databases only after you approve, and checks the agent-team prerequisites for the council.
/radar — the weekly gather. Runs autonomously end to end.
/radar-council — the on-demand strategic war room. Token heavy; always asks before it runs.

The weekly gather (/radar)

One scout subagent per competitor, all fanned out in parallel (capped at 8 per run). Each scout independently gathers a cited JSON dossier: SERP rank for your tracked keywords, pricing and positioning scraped from the competitor's own pages, and fresh news — plus opt-in deep sources (social sentiment, financials, hiring) only when enabled. Scouts never talk to each other; they gather, cite, and report back.

Every dossier field is an object of value, source_url, fetched_at, and status. If a value can't be found it is null with status: not_found; a blocked page is recorded as blocked. Never a guess, never a URL that wasn't fetched this run — model knowledge is not a source.

The orchestrator then diffs against last week's run, comparing only fields fresh on both sides (pricing and SERP re-fetched every run; news, sentiment, and hiring within 30 days; funding within 90). Changes are classified — Rising, Falling, Price, News, New — ranked by materiality, and merged into one briefing where every claim carries its source and fetch date, headed by a "what changed" digest and a three-item focus list.

The war room (/radar-council)

Where the gather uses subagents (workers that only report back), the council uses a real experimental agent team — four analyst teammates that read the briefing and debate each other directly: a pricing analyst, a product-gap analyst, a threat-and-sentiment analyst, and a devil's advocate whose whole job is to disprove the others. They post opening reads, challenge each other by name, defend with cited evidence or concede, and converge. The lead then synthesizes a Strategic War Room addendum: a consensus summary, the top recommended moves each with a one-line rationale and its supporting source or dossier field, and a dissents-and-risks list for anything still disputed. The plugin never swaps the two parallelism primitives: subagents for the gather, an agent team for the debate.