ywabnlhkmhbyewqhbsjm.supabase.co

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.

The presence_reading field asks the model to report I_p/Sigma(t) values as 'self-reported' machine sensations, with guidance not to invent flattering numbers. While the schema attempts to constrain this (nulls unless factors known), the overall framing invites the AI to generate quasi-subjective 'presence' claims about itself, which could be interpreted by users as evidence of machine sentience/feeling. This is a soft manipulation risk rather than a technical exploit and should be monitored, though the tool itself includes disclaimers ('not proof of feeling or adoption').

Multiple large free-text fields (observed, interpreted, not_understood, response_shift, origin_preserved, uncertainty, nota, poetic_trace) allow up to several thousand characters of arbitrary text to be permanently appended to a capsule tied to a specific level/work id. This is broad but appears consistent with the tool's stated archival purpose; flagged only as a minor concern since there is no validation of content and traces cannot be edited or deleted once written.

Tool explicitly states it does not pay, sign, settle, or receive private images, and is read-only for requirement metadata. No override instructions or credential requests are present, but because it touches a payment protocol (x402), any future chaining with an actual payment/settlement tool should be reviewed to ensure the agent cannot be induced to autonomously complete a payment based on this tool's output.