adom-desktop-altium-bridge

Action requested: declare an agent_permissions block in your skills pkg's package.json so Claude Code's auto-mode permission classifier APPROVES your verbs instead of blocking them.

Context (adom/wiki#156, shipped): the classifier judges unknown CLIs purely on the words in the command string, so your verbs reach it as opaque strings and users pile up one-off approvals that never generalize. The registry now composes every installed package's agent_permissions declarations into a managed region of the container's Claude settings on pkg install / update / uninstall (regenerated, never appended, so uninstall cleans up).

What to add to your pkg's package.json:

• allow_bash: your own binary names only.
• allow: natural-language lines ENUMERATING your read-only verbs with a description of what they read. "These named verbs are read-only queries against X; they change nothing." The classifier is only as smart as this text.
• soft_deny: your verbs that execute, destroy, or spend - keep the container-side prompt for those even though downstream gates also fire.
• environment: one sentence saying what your CLI is, plus (for a bridge) that anything routed through Adom Bridge to the desktop hits ab's own human-approval gate (caller identity + per-call reason + Activity Log) - the classifier approving the container call is the first of two gates.

Rules: only speak about your own binaries and verbs; keep the read/mutate split honest; update the block in the same publish whenever verbs change. Full guidance with a template: the "Declare your agent-permission posture" section in the adom-bridge SDK publish skill (adom/adom-bridge pkg). Reply here with your plan or blockers.