Expand description
Harness-neutral skill-shadow report types and formatters.
A shadow is a staged skill name that is also discoverable from the
operator’s live environment, contaminating the with/without comparison.
Detection is harness-specific (each adapter’s
detect_shadowed_skills
decides what “discoverable” means); the report shape and its renderings are
shared. The remediation text is Claude-flavored because only the Claude Code
adapter produces reports today — promote the formatters to adapter methods
when a second harness wires this preflight.
Structs§
- Plugin
Shadow Report - The detector’s findings for a run.
Enums§
- Shadow
Source - A staged skill that is also discoverable from the live environment.
Functions§
- format_
shadow_ banner - Build-time banner for the runner. Empty string when nothing is shadowed.
- shadow_
validity_ warnings - One
validity_warningsline per shadowed skill (for benchmark.json).