SKILL: Bridge Safe Command Advertising

Domain: command-architecture Trigger: Use when a bridged or remote client receives an announced slash-command list and that list must stay aligned with the commands the runtime will actually execute for bridge-originated input. Source Pattern: Distilled from reviewed remote-control, bridge transport, and capability-advertising implementations.

Core Method

Filter the command list before serializing any remote capability announcement, and filter it with the exact same bridge-safety predicate the runtime will use when the remote client later invokes a slash command. Let the shared serializer keep its generic user-invocable check, but pass it a prefiltered list that already excludes bridge-unsafe commands. This keeps the announced surface honest: every command the client sees is one the bridge path can actually execute, and every command the bridge path rejects stays undiscoverable instead of becoming a guaranteed dead end.

Key Rules

• Reuse the runtime bridge-safety predicate at announcement time instead of inventing a second listing-only filter.
• Apply the bridge-safe filter before generic serializers flatten commands into plain names; once only names remain, execution-policy drift is harder to spot and correct.
• Keep shared message builders transport-agnostic: they may still filter out non-user-invocable entries, but transport-specific safety should be decided by the caller that owns that transport.
• Treat advertised command lists as part of the execution contract, not as best-effort discoverability metadata.
• When bridge execution rules change, update the shared predicate so announcement and invocation stay synchronized automatically.

Example Application

When a CLI session opens a phone companion over a bridge and emits system init, filter the command registry through is bridge safe command before calling the generic init-message builder. The serialized slash_commands list then includes summary and compact, but omits local-only UI commands such as model that the bridge path would reject later.

Anti-Patterns (What NOT to do)

• Do not advertise every user-invocable command and rely on runtime failures to teach remote clients what is unsupported.
• Do not maintain a separate announcement-only allowlist that can drift away from the real bridge execution gate.
• Do not hide transport-specific safety inside the generic serializer when different callers may need different execution contracts.