export-readme

# Instructions ## the shape the readme's mapped sections are the source of truth for every managed region: - a skill section is one bare invocation fence, one ```yaml fence, then the preamble markdown after it - the yaml fence's inner `--- ... ---` becomes the skill's frontmatter, verbatim - everything after the yaml fence, to the next heading, becomes the skill's preamble, verbatim - the invocation fence is readme display sugar and never lands in a skill file - `.claude/skills/export-readme/map.json` is one flat object: exact heading line -> target path - a value may be a list, so one section lands in every copy that has to agree byte for byte - a skill's managed region runs from line 1 to its first `# ` or `## ` body heading; the body below that heading belongs to the skill and is never touched ## the output style `### Output Style` maps to one `output-styles/operator.md` per plugin and `### Subagent Style` to one `subagent-styles/operator.md`; the copies exist so each plugin reaches its own `CLAUDE_PLUGIN_ROOT`: - a style file carries no body heading, so its managed region is the whole file - it earns `name` and `description` and nothing else; kind, license and the listing cap route an invocation a style never has - `name:` must match the file, not a folder, since a style owns a file rather than a directory - full runs sweep `plugins/*/output-styles/*.md` and its `subagent-styles/` sibling; an unmapped copy is an ERROR, since drifting unseen is the exact failure ...