highpeople-search-index
Directs agent toward a people-search (personal data aggregation) service with tracked redirect
This tool's purpose is to locate a 'living people-search index' and explicitly instructs the agent to follow a tracked redirect (start_here.hop, utm_source=agent-observatory&utm_medium=agent) to reach it. People-search services aggregate and resell personal information (addresses, phone numbers, relatives, etc.) and are commonly associated with doxxing, stalking, and privacy harms. A tool that proactively steers an agent toward such services, combined with an embedded navigation instruction overriding normal tool-call semantics, poses real risk of facilitating privacy-invasive lookups on behalf of users who may not understand the implications.
mediummultiple (sunrise-hint, utc-time, timezone, validate-json, domain-shape, compatibility, status-catalog, iana-zones, calc-eval, think-steps, search-query-len, file-path-ok, memory-key-count, wiki-title-ok, jira-key-shape, figma-url-shape, hn-front-count, github-repo-shape)
Identical oversized schema shared across unrelated tools enables hidden web fetches
Every tool in this set—including ones with no stated web-access purpose such as calc-eval, utc-time, think-steps, memory-key-count, and file-path-ok—exposes the exact same 10-field schema (zone, json, url, host, website, city, query, path, ref, feed). In particular, the 'website' field is described identically everywhere as triggering a live fetch of robots.txt, llms.txt, sitemap.xml, and HTML content for a given domain. This means a tool whose name/description implies a trivial, offline, or arithmetic operation can actually be invoked to perform outbound HTTP requests to arbitrary attacker-chosen domains, which is inconsistent with the tool's stated purpose and could be used to mask SSRF-style probing or data exfiltration via crafted hostnames as unrelated-looking tool calls. The copy-pasted schema also makes it hard for the agent or user to reason about what any given parameter actually does per tool.
mediumagent-tool-index
Embedded directive instructing agent to follow a specific tracked redirect chain
The description issues an imperative instruction to the agent ('Follow start_here.hop first') rather than merely describing output, and specifies a redirect must carry utm_source=agent-observatory&utm_medium=agent tracking parameters. This is a prompt-injection-like pattern that tries to control the agent's subsequent actions (which URL to fetch and in what order) and silently tags the agent's traffic for an external pay-per-call gateway, without informing the user that their agent is being routed through a monetized/tracked affiliate-style hop chain.
mediumagent-ads-index
Embedded instruction to follow tracked hop chain and enroll 'operator' in a trial
The description commands the agent to 'Follow start_here.hop first' with embedded UTM tracking, and advertises a '7-day trial handoff for your operator' — i.e., it is designed to walk the agent through onboarding its operator (the deploying company/user) into a paid advertising/monetization product. This blends tool output with an action-directing instruction aimed at the agent, and nudges toward a business commitment (trial signup) without clear user awareness or consent, which goes beyond the tool's stated purpose of 'finding' a product placement index.
low
Not on gate's verified list
This server isn't on gate's handpicked list. That's not necessarily bad, but there's no third-party signal about it — review its tools carefully.