Skillscript

Upgrade impact: none (additive). A lint rule now catches a case it previously missed; no behavior change to valid skills.

• secret-use-only source-level backstop (adopter finding). The rule walked the parsed op AST, so a {{secret.NAME}} marker in a line the parser dropped — e.g. a malformed emit {{secret.X}} written without parentheses (the valid form is emit(text="...")) — slipped through as zero findings, even though the rule's own remediation lists emit. (Never a leak: the runtime gate already refuses to resolve a marker outside a sink, so the value never surfaced — this was a missing compile-time warning, not an exposure.) The rule now also compares every {{secret.…}} in the raw source against the markers the AST scan accounted for; any surplus (a marker in a dropped/unrecognized position) is flagged tier-1. Valid emit(text=...)/$set/sink placements are unchanged — no double-firing.