What it does. Codex metadata for the skill: display name “Retro”, short description “Conduct a retrospective on a coding session.”, and policy.allow_implicit_invocation: false, which stops Codex from starting the skill on its own.
Why it’s here. Codex reads it for its skill picker. The repo’s .agents/invocation.md says every skill carries one and that “It holds Codex UI metadata”. The policy line matches disable-model-invocation: true in SKILL.md, because that file also says “Keep the two in sync: a skill is user-invoked in both harnesses or neither.” SKILL.md itself does not refer to this file.