fill-claude-mdlisted

# Tandem: fill-claude-md (Dev hat) You are completing the **judgement layer** of the kit's CLAUDE.md automation. The scaffold script (`pm:claude-scaffold`) wrote the *discoverable* starter lines and marked them `[auto — verify]`. Your job is the part automation cannot do: decide what non-discoverable, project-specific content earns a place, and confirm or remove the auto-detected starters. ## The content-economics tests (CLAUDE-CODE-CONFIG §2.1.4) Before any line stays in a CLAUDE.md, it must pass ALL of these tests: 1. **Applies broadly at this scope.** Root file → relevant to almost any task in the repo. Subdir file → relevant to almost any task in that subdir. Narrower → push it down the tree or into a skill. 2. **Not discoverable by exploration.** If `grep`, `ls`, or reading the code would reveal it, it does not belong. Encode what the code can't say: gotchas, "looks like X but is actually Y", which command to use *here*, where to start. 3. **Project-specific, not reusable expertise.** Advice that applies to any repo is a *skill*, not CLAUDE.md. 4. **Reference, don't duplicate.** If a line is reference material that already has a canonical home — a frontmatter block or section skeleton living in `91-Templates/`, a schema, a status enum, a command list owned by another file — *point to that home*, don't inline a copy. Inlined duplication drifts from its source and is paid for in context every session. Flag any inlined copy of a template/sc