WWTDigital Skills

brand-scan

draftv0.1.1in WWTDigital Brand Scan

Run or read a brandscanner assessment of a company's digital presence: security posture, tech stack, app-store reviews, public financials, email-auth records, benchmarks against the scanned cohort, and the resulting scorecard, using the brandscanner MCP server bundled with this plugin. Use when someone says "scan this brand", "run brandscanner on <company>", "what's <company>'s scorecard", "how does <company> compare to its cohort", "pull the tech stack for <company>", or "prep a digital assessment of a prospect". Not for general desk research without the tool, for writing the pitch deck or proposal that follows, or for anything about WWT's own internal systems.

Brand scan

Runs, refreshes or reads a brandscanner assessment of one company ("brand"): evidence scans of its website, email domain and mobile apps, public financials, six research dimensions scored 0-10, and a saved scorecard with version history and percentiles against the cohort of brands already scanned (currently quick-service restaurants). For sellers, strategists and researchers preparing a digital assessment. Output is a short readout in chat, backed by the scorecard saved in brandscanner. Per-tool notes and the scorecard JSON shape are in references/tools.md.

Inputs

  • The company name, from the person. Never infer or pick one.
  • Which parts they want: a read of what exists, a full run, or specific scans (see step 4).
  • Optional: website URL, app-store IDs or ticker if the person knows them (resolve_brand usually finds them).

Steps

  1. Connector check. The brandscanner server connects when this plugin is installed, but it needs a one-time sign-in. If tools like list_brands and get_scorecard aren't available, brandscanner isn't connected: tell the person to run /mcp in Claude Code (or accept the sign-in prompt in the Claude app), sign in, and ask again. Then stop.
  2. Read before you scan. Reads are cheap; scans are slow and hit external services. Call get_scorecard for the brand (and brand_benchmarks if they asked about the cohort).
    • If a scorecard exists, show the person its scan_date, scores and history_count, and ask whether the existing one answers their question before starting a new run. A read-only request ("what's the scorecard", "how does it compare") ends here, at step 9.
    • If status is not found, the brand is new: say so and go on.
    • Call scan_status for the brand. A run that is still running with no finish time long after it started is an orphan; close_stale_scan_runs (default 60 minutes) marks it timeout. If a previous save_scorecard ended in persist_failed, call resume_scorecard_save with the brand slug (from list_brands) instead of re-scanning; no_checkpoint means there is nothing pending.
  3. Confirm the target, then resolve it. Scans touch the company's live systems, so before the first one restate the company and get a yes. Then resolve_brand with the name; it returns website URL, iOS/Android app IDs, ticker, social handles and competitors. If the result looks like a different company (wrong domain, wrong sector), stop and ask.
  4. Agree the scope. If the person didn't say, ask which of these they want, with rough timings (estimates, not measured):
    • Website: scan_security (headers, TLS, cookies, dead links, policy pages) and detect_tech_stack (frameworks, CMS, CDN, analytics, payments), well under a minute each.
    • Email domain: check_dns_auth (SPF, DKIM, DMARC), seconds.
    • Apps: analyze_app_reviews (ratings and recent reviews, both stores) and track_app_version (iOS version and release cadence), a minute or two; they scrape the stores.
    • Financials: fetch_public_financials (market cap, revenue, margins); seconds, returns status: private for private companies and can be rate-limited.
    • Breach history: search_breach_db; optional, no longer feeds a scored question. A full run is usually 5-15 minutes end to end. Default to everything except breach history.
  5. Open the run. start_scan_run with the brand name; keep the run_id. Everything from here to step 8 belongs to that run.
  6. Run the scans from step 4 with the URL, domain and IDs from resolve_brand. Keep two lists as you go, tools_succeeded and tools_failed (with the reason, e.g. fetch_public_financials (rate_limited)). A failure in one scan does not stop the others.
  7. Score and save.
    • score_research with the brand name and this run's fetch_public_financials, analyze_app_reviews and detect_tech_stack results passed as financials, app_reviews and tech_stack. Omitting them scores those dimensions as flat 5.0 stubs. Pass any web research the person supplied as web_research_text.
    • apply_overrides for the brand, so analyst overrides survive the re-scan (its description asks for it after scoring and before the final save).
    • save_scorecard with scorecard_json built as references/tools.md describes and merge left true, so categories this run did not assess keep their previous scores. Keep the returned scan_id. If it returns persist_failed, the result is checkpointed: tell the person, and note that resume_scorecard_save replays it later.
  8. Close the run. finish_scan_run with the run_id: on success status: ok, the scan_id, and the two tool lists; on failure status: failed with error_details {stage, message}. Always close it, including after a failure, or it shows as stale.
  9. Verify and present. Call get_scorecard again and check: scan_date is today (for a fresh run), the metadata.tool_results keys match the scans that ran, and every section the person asked for is present with a non-stub value. Then present, in this order: overall, core and research scores with deltas from the previous scan; the category benchmarks (score, percentile, cohort average) from brand_benchmarks, plus the cohort rank; the research dimensions with their evidence; and a plain list of scans that failed or returned nothing, with the reason. Repeat any rubric_mix_warning the server returns, and say when a brand is outside the cohort's sector, since percentiles then compare unlike things. If the person asks where WWT could help, map_solutions maps low-scoring sub-themes to WWT capabilities; offer it, don't run it by default.

Output

A chat readout of the scorecard as in step 9: numbers first, evidence second, gaps last, with the scan_id and scan_date so the result can be found again. No file unless asked; if the person wants a document, hand the verified numbers to a presentation skill rather than retyping them.

Guardrails

  • Confirm the company before the first scan, and never scan one the person hasn't named.
  • Never present partial output as a complete assessment. The core category scores come from the full monthly pipeline (crawls, Lighthouse, accessibility checks) that this server does not expose; a run from here refreshes the evidence tools and research dimensions and, with merge: true, carries the rest forward. Say which is which.
  • Scorecards can include third-party breach data and scraped reviews. Keep them internal unless the person says otherwise, and don't paste raw breach records into shared channels.
  • Don't invent scores. A dimension marked stub or model_prior is a placeholder; call it that.
  • Always pair start_scan_run with finish_scan_run; never call close_stale_scan_runs while a run you know about is still in progress.

Examples

User says: "Run brandscanner on <brand> and tell me where they're weak." Claude does: Checks the tools are available, reads any existing scorecard and asks whether a fresh run is wanted; confirms the company, resolves it, agrees the scope, opens a run, runs the scans, scores, saves, closes the run, re-reads the scorecard and presents the lowest percentile categories with their evidence and the list of scans that didn't return data.

User says: "How does <brand> compare to its cohort?" Claude does: Calls get_scorecard and brand_benchmarks only; presents rank, category percentiles and cohort averages with the scan date, repeats the rubric-mix warning if present, and offers a re-scan if the scorecard is old. No scans run.

Should NOT trigger: "Write a two-paragraph company overview of <brand> for the intro slide." That's desk research and deck writing; use a research or presentation skill without brandscanner.