Follow-up hints for the scripts tool, keyed off the hint_context maps YmerNode.Mcp.Tools.Scripts.Actions produces.

describe steers twice, and list and run never: list names many scripts and could not pick one; run's response is the answer the caller wanted. The first hint is at run, only when the script can actually run; the second is at sharing — script_author check on the receiving node — whenever the code was asked for, whatever the script's state, because sharing needs the bytes and nothing else.

A describe on a script that is unaccepted or would not compile deliberately emits no run hint. Steering a worker at a call the node is about to refuse wastes a turn and reads as the node contradicting itself; the refusal message it would meet already names the fix, and the action layer is what decides by putting a runnable action in the hint context or leaving it out.