Problem
The skill-effectiveness signal window (window_size=50) includes invocations of commands that were retired mid-window (flow:judge, flow:author, flow:status, flow:board, crew:board — retired in the 2026-06-19 command-surface simplification). Their historical invoke_count values appear as high-invoke / zero-ratio entries alongside active skills, so the retro analyst cannot distinguish a skill that fires but does not help from a skill that was retired during the window.
Suggested direction
When a skill folder is deleted (command retired), write a retired_at tombstone into telemetry. computeSkillEffectiveness then filters or annotates any per_skill entry whose retired_at falls within the window, giving a clean signal over active skills only.
Source
Raised by the retro-analyst during a cycle-end retro (maintainer-inbox finding 01KVEPRYP8HBFZJEKS6RAGP223).
Problem
The skill-effectiveness signal window (window_size=50) includes invocations of commands that were retired mid-window (flow:judge, flow:author, flow:status, flow:board, crew:board — retired in the 2026-06-19 command-surface simplification). Their historical invoke_count values appear as high-invoke / zero-ratio entries alongside active skills, so the retro analyst cannot distinguish a skill that fires but does not help from a skill that was retired during the window.
Suggested direction
When a skill folder is deleted (command retired), write a retired_at tombstone into telemetry. computeSkillEffectiveness then filters or annotates any per_skill entry whose retired_at falls within the window, giving a clean signal over active skills only.
Source
Raised by the retro-analyst during a cycle-end retro (maintainer-inbox finding 01KVEPRYP8HBFZJEKS6RAGP223).