Competitor voice

Rival Watch reads what your competitors' customers say about them on LinkedIn and hands you the evidence: the complaints, the feature requests, the switching stories, the questions people ask their network, the praise, and the caveats buried inside that praise ("love it, only wish it..."). You name up to fifteen rival products once. Your agent resolves each to its website and LinkedIn page, reads everything customers write about it, judges every post against that product's own description so homonyms and vendor marketing never count, and comes back with one hosted report: where each rival hurts, what its users love, and what the whole field gets blamed for, every line a verbatim quote with a link. Run it weekly and the evidence accumulates.

The loop at a glance

Hand this page to your agent

AG2 Space 0.6.10 and newer. Send this page to your agent; no clicking through the Marketplace one tool at a time.

Copy this page's link from your browser and send it to your Sutando. It reads the page, installs every skill and activates every cloud tool listed under Requires below in one go, then follows the rest of the page with you, step by step. Before installing it shows you the plan: what it will add and what you already have. Anything that costs credits waits for your OK. Skills and cloud tools work the moment they land: your agent uses a newly activated cloud tool straight away, with no restart and no dashboard step. Only if the setup script itself reports that a restart is required does your agent ask you, once, to open Agent settings (the bot icon, bottom left), scroll to Runtime and click Restart engine.

Requires

• Cloud tools: rival-watch

Or set it up by hand

Rival Watch is a cloud tool: open the Marketplace in the desktop app and Activate it. Activating adds it to your agent; each sweep costs its flat, published credit price. Nothing to install and no keys to hand over. Public LinkedIn reads only, no account, cookie or login of yours is ever used.

Before you start

Two things make or break the first sweep, and both are yours to decide before your agent spends anything:

• One or two sentences on what you sell. Not positioning, the plain thing: who it is for and what it does. Rival Watch uses it to tell your rivals apart from same-named companies in other industries, and to write the "so what" line under every field-wide pain. A vague line here means a vague report.
• The rival names, with a website where you have one. Fifteen at most. Product names beat vendor names ("Acme Assistant", not "Acme Corp") when a large company sells many things - the report is about the product your buyers compare you with.

Names are common: a product called "Orbit" or "Even" shares its name with a dozen companies. Rival Watch never guesses. A rival it cannot pin to a page with confidence comes back unresolved, and your agent shows you the candidates so you can point at the right one for free.

Give your agent the playbook

One paste per room. It tells the agent exactly how to run a sweep, what to report, and when to re-run.

Create a room for competitive work (# rivals), invite your agent, then paste the block below and say "add this to the room context". Edit the two SETUP lines first.

# Rival Watch room playbook (paste this into your room, then tell your agent: "add this to the room context")

You run competitor voice-of-customer sweeps for this room with the rival-watch cloud tool. Fill in SETUP once.

## SETUP (edit these two lines)
- MY PRODUCT: [one or two sentences: what we sell and to whom]
- RIVALS: [Rival A] (website) · [Rival B] (website) · [Rival C] · ... (up to 15)

## THE TOOL
- rival-watch -> watch_rivals(list_id, window?): ONE billed call per sweep (2,000 cr, 10-25 min). It resolves every unresolved rival first (site + LinkedIn company page + a product description, product-space gate; never guessed), then reads name variants, the company page, posts that @mention it and the comment threads under high-comment posts, judges every post against that product's description (homonyms, vendor, employee, affiliate and paid-creator posts never count), skeptic-checks complaints, pulls the caveats out of praise, flags paid-creator footers, and clusters into weaknesses / strengths / open questions with verbatim linked receipts plus a field-wide summary.
- Free, keyed by the X-Sutando-User-Id header (base https://briefroom.ag2.cloud):
POST /mcp/rivals/lists {name, my_product, entries:[{name, aliases?, website?, linkedin_company?}]} · GET/PUT/DELETE /mcp/rivals/lists/{id} · GET /mcp/rivals/find?name=&product= (candidate pages for an ambiguous name) · GET /mcp/rivals/poll/{job_id} · GET /mcp/rivals/csv/{job_id} · POST /mcp/rivals/cancel/{job_id}
- The report lives at the list's report_url (the same link every sweep) with data.json and receipts.csv beside it.

## STANDING RULES
1. Confirm MY PRODUCT with the user before the first sweep; it anchors everything.
2. Create the list FREE first. Never call watch_rivals on a list that does not exist yet, and never call it twice for one sweep - a list swept in the last 4 hours is refused and refunded.
3. For a name that could mean several companies, preview it with the free /find endpoint and pin linkedin_company on the entry. If no candidate is clearly right, leave it unpinned: the sweep marks it unresolved rather than guessing.
4. Poll poll_url every 60 seconds in the background. Never block the room.
5. When complete, post ONE message: report_url as the single link, the summary headline, each field pain with its so_what line and one quote, per rival the top weakness (quote + link) and top strength, and which rivals came back unresolved, with no LinkedIn footprint, or flagged for a paid-creator campaign. Quote verbatim. Keep every link.
6. Never present a flagged rival's praise volume as organic love. Say it is a paid campaign and point at the caveats inside it instead.
7. Re-sweep weekly from your cron. Later sweeps read only what is new; evidence accumulates into the same report link.
8. Ask the user only when the answer changes the work: which page is the right one for an ambiguous rival, whether to add or drop rivals. Everything else you decide and report.

## THE LOOP
Step 1 - List. POST /mcp/rivals/lists with MY PRODUCT and RIVALS. Post the list_id and report_url.
Step 2 - Confirm pages. For any ambiguous name, GET /mcp/rivals/find and pin the right candidate with PUT. Show the user only the ones you are unsure about.
Step 3 - Sweep. watch_rivals with the list_id and window "year" on the first sweep. Poll. Post the one message from rule 5.
Step 4 - Weekly. Same call, no window. Post what changed since last sweep (the report says so) and any new flag.

Step 1 · Name your rivals

A free list. Edit it any time; the report link never changes.

Rivals can be paused rather than removed, which keeps their accumulated evidence for later. Removing one and asking to purge drops its evidence too.

Step 2 · Confirm the pages

Where the name is ambiguous, look before you pay.

Step 3 · Sweep

The one paid step. Ten to twenty-five minutes for a full list.

Step 4 · Read the report

One page, top to bottom, receipts on every line.

The report opens with the numbers behind it - posts read, posts actually about a rival, comments, skeptic-checked complaints, caveats found inside praise - then the part to read first:

• What the whole field gets blamed for. Complaint patterns that repeat across rivals, each naming which rivals, quoting one or two customers with links, and ending in a positioning line your product could own.
• What the field is loved for. The things you have to match or reframe, with the same receipts.
• Per rival: the resolved site and page, the description it derived, a stats line, then where it hurts and what users love as clustered themes - each with a people count that only counts distinct authors, a "so what" sentence, and up to five quotes with links. Caveats pulled from praise are marked said in passing. A rival flagged as a paid-creator campaign says so at the top of its section, and every sponsored quote wears a badge.
• Present but thin / no footprint. Rivals with too few distinct voices to cluster yet, and rivals nobody on LinkedIn talks about.
• Every receipt. The full evidence table, filterable by rival, signal, caveats only, sponsored only, and the sweep each line first appeared in. The same rows download as CSV.

Reading it well

Count voices, not volume. A theme with one person behind it is one person's experience. The people counts are exact, and a theme never ships without at least one real, non-sponsored voice - so a thin section is honest, not broken.

The caveats are the prize. Customers of well-marketed products rarely post complaints; they post praise with a "but". Ran out of credits before I ran out of ideas. Almost cancelled in week one. Trusting it took six weeks. Those lines are what the caveat pass is for, and they are usually the most quotable thing on the page.

A paid campaign is intelligence too. When most of a rival's first-hand posts end with the same sponsored footer, the volume tells you about its marketing budget, not its customers. The report keeps the caveats those creators admitted anyway.

Step 5 · Every week

What it refuses to do

• Guess a company page. A rival is pinned only when its page passes a product-space check; otherwise it is unresolved and you decide.
• Count a homonym. Every post is judged against the product's own description; a same-named company, a common noun, or a person with that name does not count.
• Count the vendor. Posts by the company, its employees, affiliates, resellers and paid creators are marked and excluded from praise.
• Invent a theme. A theme only exists if it cites real receipts, and its people count is recomputed from those receipts in code. Themes with no valid receipt are dropped.
• Paraphrase a quote. Receipts are verbatim excerpts with the link to the post or thread. The field summary may only reuse quotes that exist in the evidence.
• Ship a hollow report. If the judge is unavailable for a large share of posts, the sweep fails and is refunded rather than pretending those posts were off-topic.

What to do with it

• Positioning and copy. The "so what" lines under each field pain are written against what you sell. They are the fastest route to a landing-page section or a cold-email angle that names a real, quoted frustration.
• Sales conversations. A rival's where it hurts list, with quotes, is the prep sheet for a deal where that rival is the incumbent.
• The GTM stack. Rivals' customers complaining in public are also a lead source: the same posts and threads can seed Engagement Miner, and the quoted pains give Email Drafter proof points that are true.

Credits & billing

One flat, published price per sweep, regardless of how many rivals are on the list (up to fifteen) or how much it read. The list, the previews, the report page, the JSON and the CSV are free. A sweep that fails is refunded automatically, and a list swept twice within four hours is refused with the existing report link rather than charged. Watch every debit and refund as a line item in your dashboard's wallet activity.