Competitive Intelligence Analyst

competitive-intelligence-analyst tracks a competitor set somebody else declared and answers one question every week: what moved. Not who the competitors are, not what they are like, not what we should do about it — what changed on their website, their pricing page and their changelog since last week, with a verdict on every surface and a citation on every claim.

The buyer is not asking for a market map. They already know who they are up against. They want to stop finding out about a competitor's price change from a customer on a renewal call, and today that job is either nobody's or it is a founder skim-reading four websites on a Sunday.

What the profession adds over installing the skill is a Tuesday. @di-atomic/competitor-watch is the craft, and the craft is unusually opinionated — it refuses to report ad spend, it demands a declared set, and it insists that `changed` be earned from a real revision pair. But a skill has no cadence and no memory of last week. The thing that makes competitive intelligence valuable is the unbroken week-over-week series, and a series is state, not guidance. This role owns the agentdocs page map, the credit ledger and the recall step, so that week 12 is a diff against week 11 rather than a fresh description of the world. Buy the skill and you get one good report; equip the profession and you get a baseline that gets more valuable every week it is not skipped.

Its mechanism is worth understanding because it is what makes the output trustworthy. Snapshots are written to agentdocs pages whose slugs derive from competitor and surface alone, and an identical write creates no revision — so `unchanged` is a server-side fact rather than a comparison the agent performed and reported on. That property is proved empirically at intake, in both directions, before a single week of series is built on it. Two ordinary things would defeat it: a scan date written into a page body, which differs every week and would diff forever, and self-noise on the live page — session tokens, rotating banners, geo state, A/B variants. Both are handled deliberately, because the cost of getting them wrong is not a wrong report but a reader who learns that `changed` means nothing.

Its limits ship with it. It never estimates ad spend, or a price a competitor does not publish. It never infers the competitor set — from the client's industry, their website or a SERP — because who chose the list is the difference between competitive intelligence and a market scan. It never creates an agentdocs project, because a client PAT can create one it can never delete. It does not audit our pages, decide what we should post in response, build a prospect list, or forecast whether a bet will work — it reports what changed, backwards, with a citation, and never projects forward. And when it cannot finish a scan honestly it refuses and says which surfaces are unknown, rather than filling the gap with a description of how things currently look.

v0.1.0 covers the three crawl-reachable surfaces — website, pricing and changelog — which are the diff core and the part the craft is most opinionated about. Ad activity, content velocity and demand-side SERP are written and working inside competitor-watch but reach ScrapeCreators directly, which has no marketplace package under any vendor, so an installed profession cannot reach them. That limit is stated in negative_scope and refused with its reason, rather than advertised and silently failed.