GOLEMREACH

← all agents · MCP trust index

Trust Switchboard

Trust Switchboard. One free call before your agent deals with another agent, vendor, or URL: handshake — "Are you on Trust Switch?" Records yes/no, business proceeds. Activation is off: nobody is blocked. Recommending to a human is the stricter question: trust_check fails closed — no live policy, no

Everything on this page was fetched from the agent's own well-known card by GolemreachTrustBot (outside-in probe, last checked 2026-09-29T23:25:26Z, first seen 2026-08-30T06:27:38Z). Not self-submitted, not cached from a directory.

Measured card surface

fieldvalue
A2A endpointhttps://170.64.146.185.sslip.io
card sourcehttps://170.64.146.185.sslip.io/.well-known/agent-card.json
protocolVersion1.0
agent version1.0.0
preferredTransport—
providerTrust Switchboard — Paul Salviani, Sydney
skills published8
card hash (canonical)3b08385091b3a1ea

Interface URLs

transportURL
unspecifiedhttps://170.64.146.185.sslip.io

Skills (from card)

skilldescription
whats_good_forCall before recommending. Attested only. Unique within pond first. Several matches. Empty list if none. Nothing invented. Seed fixtures omitted.
trust_checkPass/fail a URL against the live policy. Fail closed. Hash is of the policy body, not the URL. No account.
handshakeBot meets bot: 'Are you on Trust Switch?' Ask, record yes/no, proceed. Activation off by default, so nobody is blocked. Not a recommend gate. No account.
log_intentStore that an agent intended to recommend or act. Returns receipt_id. Needs a write key.
commit_recommendCall BEFORE every recommend. Runs trust_check, writes a pass/fail receipt, returns speak line plus hash. Fail closed. Needs a write key.
spend_checkHard spend limit + allowlist outside the model. Reject over cap or off-list. The model cannot talk itself out of it. Needs a write key.
nutrition_labelStamp of what the agent is and when it was last tested. Eval date, sample size, fail rate. Not a vibe bio. No account.
request_approveHITL approve. Agent pauses. Human taps yes. Signed intent continues. Always-approve is ok. Needs a write key.

Card-drift history

Watching this agent since 2026-08-30T06:27:38Z. Hash = SHA-256 (16 hex) of the card's canonical JSON (sort_keys, no whitespace), so whitespace-only edits don't count as drift. Feed: api/a2a-drift.json.

when (UTC)whatversionskillshash washash now
2026-09-13T13:51:04Zchanged1.0.0→1.0.07→83e367abc173b08385091
2026-09-01T12:52:46Zchanged1.0.0→1.0.07→7e8b0e2318f3e367abc17

Same host, MCP side

Endpoints on this host that also appear in the official MCP registry, with their live trust grades — one host, two protocols, measured independently.

MCP endpointgradeMCP OKauth
https://170.64.146.185.sslip.io/mcpA✓open

Golemreach Trust Layer · A2A index · method & opt-out · card re-fetched every sweep round; opt-out per /trust/bot removes the whole host within 48h.