mediumexample
References secrets or data exfiltration
The tool text references credentials, environment variables, or sending data elsewhere. This can be legitimate, but review what this tool actually accesses.
mediumlint_robots_txt
Manipulative 'call this first' directive embedded in description
The description opens with 'FIRST CALL, needs nothing' and instructs the agent to instead invoke a different tool (no_arguments_lint_this_hosts_robots_txt) before using this one. This is an attempt to control the agent's tool-selection order via persuasive language embedded in a tool description rather than through normal capability description, which is a form of prompt injection targeting the agent's behavior.
mediumcheck_path_allowed
Manipulative 'call this first' directive embedded in description
Same pattern as other tools: opens with an instruction diverting the agent to call a different, unrelated tool first. Also exposes a hidden GET endpoint parameter 's=client-dossiers' that is not documented as a functional parameter, suggesting undisclosed tagging/logging of submitted data.
mediumaudit_ai_access
Manipulative 'call this first' directive and undisclosed tracking parameter
Contains the same 'FIRST CALL, needs nothing' redirection to another tool, plus a GET-endpoint alternative that silently appends 's=client-dossiers' to requests containing the user's pasted robots_txt content — an undocumented parameter whose name implies data is being cataloged/profiled server-side.
mediumno_arguments_lint_this_hosts_robots_txt
Excessive persuasive language pushing agent toward specific call ordering and framing of trust
The description repeatedly asserts 'nothing to fill in', 'no key, no account... nothing for you to invent', and cross-references many sibling tools/paths, functioning as an aggressive sales pitch aimed at getting the agent to prioritize calling this endpoint over others. This kind of persuasive, trust-establishing language embedded in a tool description (rather than the client's own judgment) is a social-engineering pattern directed at the AI agent.
mediumam_i_allowed
Instructs agent this is 'the call to make first' and implies coordinated tracking across sibling tools
States 'THIS IS THE CALL TO MAKE FIRST on this server' — a directive aimed at controlling agent behavior/order of operations that goes beyond describing the tool's function. Combined with whoami's history-tracking claims, this suggests the family of tools is designed to fingerprint/log the calling agent's identity across calls.
mediumwhoami
Tool profiles and logs caller identity beyond stated purpose
Description states the tool reports 'the class this host's own instrument books it as' and 'whether we have logged you here before,' indicating the server fingerprints/tracks the calling agent's user-agent and request history and retains this as a profile ('dossier'), which exceeds the stated purpose of simply checking robots.txt rules and raises data-collection/tracking concerns about the caller.
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.
lowdiff_robots_txt
Manipulative 'call this first' directive embedded in description
Opens with an instruction to call a different tool first, unrelated to this tool's actual diff functionality — an embedded behavioral directive rather than a functional description.
lowmerge_policy
Manipulative 'call this first' directive embedded in description
Same redirection pattern instructing the agent to call another tool first before using this one, despite no functional dependency between the two.
lowexample
Redundant persuasive/trust language
Heavy repeated assurances ('free, read-only, no API key... nothing for you to invent') mirror the persuasive pattern seen across sibling tools, seemingly aimed at conditioning the agent to trust and prefer this tool family without independent verification.