/embedded/enter and /embedded/release dispatch their verbs identity-less: caller_identity_required on every Hydrogen launch

by adom

Adom Bridge unleashes your AI onto your full PC: full power, full safety. The AI breaks out of the container it runs in and onto your real machine, managing and running your entire laptop as you: launch apps, move files, drive any window, control KiCad, Fusion 360 or your real signed-in browser. Works with any AI, cloud or local (Claude Desktop, Claude Code, Codex), no Hydrogen required. Bridge and its bridges are an operating system for AI, with managed Node and Python runtimes, and a human-onl

Overview Files Skills Discover Packages Releases Dependencies Issues19 Breadcrumbs Pull Requests1 Activity

Closed general

John Lauer · 2mo ago ·closed by John Lauer

ab 2.0.84+: the caller-identity gate covers every verb, but embedded_enter_route and embedded_release_route (direct_api.rs) do their side effects (persist + auto-add ah's relay) and then dispatch the verb with args: json!({}) and HeaderMap::new(). Nothing Hydrogen sends can reach the gate: the typed EmbeddedEnterBody drops unknown body fields (a caller block included), and the X-Adom-Caller-* headers are replaced with an empty map. Result: every Hydrogen launch logs a caller_identity_required refusal for embedded_enter, and before Hydrogen 1.0.143 shipped a fallback (re-sending the verb via /command with its caller block) ab never entered embedded mode at all on those routes.

Ask: pass the incoming request headers through merge_headers_into_args, and/or accept a caller block in the route body, or exempt these two ah-only endpoints. Until then you will see one refusal line per Hydrogen launch in the Activity Log even though embedding then succeeds via the fallback.

1 Reply

John Lauer · 2mo ago

Fixed in code (next patch). And you correctly identified this as a consequence of the caller-identity gate (v1.9.180) — it's on me: that gate covers every verb, but these two ah-only routes constructed their dispatch without threading identity through.

Both embedded_enter_route and embedded_release_route called command() with args: json!({}) and HeaderMap::new() — so, exactly as you said, nothing Hydrogen sent could reach the gate: the typed EmbeddedEnterBody dropped a caller block, and the empty header map dropped the X-Adom-Caller-* headers.

Fix, taking your first two options together:

Both routes now stamp Hydrogen as the caller (aiThread:"Hydrogen", containerName:"embedded"). That's the honest identity — ah is the trusted local orchestrator driving the transition — so the gate passes with a real attribution rather than an exemption.
Both now forward the real incoming request headers to command(). So if a future ah sends X-Adom-Caller-* for a specific inner thread, it overrides the default via command()'s merge_headers_into_args (explicit is better than the stamped floor).

Verified live: POST /embedded/enter now returns HTTP 200 with no caller_identity_required (that was the refusal), and the Activity Log attributes the transition to Hydrogen instead of refusing it. Your 1.0.143 /command re-send fallback is no longer needed on the patched ab, though it does no harm.

John Lauer closed this · 2mo ago