Wiki Maintenance & Tooling Lessons

Operational reference for anyone editing or extending this knowledge base. These are not facts about Fuller; they are the hard-won rules from building the wiki — what breaks, what works, and which conventions to follow. Synthesized from the 2026-06-27 migration session.

Editing frontmatter safely

The single most damaging class of error here is mangling YAML frontmatter with naive string tooling.

• Parse frontmatter line-based. Find the closing delimiter as a whole line (lines.index("---", 1)), and edit only the target key's lines. Never str.split("---") — article slugs and bodies contain --- (e.g. climate---architecture), and a substring split folds the body into the frontmatter.
• Never run a DOTALL .* over a frontmatter block. A regex like (?ms)^sources:\s*(\n(?:\s*-\s+.*\n?)*) with the s flag lets .* eat across newlines and silently strips every field after sources:. This stripped 78 articles in one pass during migration.
• For bulk metadata edits, round-trip through structured data. Have subagents return strict JSON ({slug, tags, aliases}), then apply changes with a deterministic line-based writer. Reserve agent direct-writes for whole-file prose (compiling articles), never for patching frontmatter in place.

Sandbox & scripting constraints

• cat and heredocs are blocked in this environment (command not found: cat; the same applies to head/tail/sed/awk). Generate files with printf, Python (open(p,"w").write(...)), or the Write tool — and verify byte counts, because a failed cat >file silently leaves a 0-byte file.
• Validate field→path conversions. When migrating bullet-field records (- **Receipt:** ...), only treat a value as a path if it is path-shaped (starts with .//../ or contains / plus an extension). Free-text values like "not owned" or "in-store cash purchase" otherwise become dangling provenance links.

Layer boundaries: raw / inventory / datasets

• Books ingest by reference, not by copy. This repo is a Wiki.js site; book directories and their ebook formats are served from the repo root and stay there. Ingest a complete book as a thin raw/repos/<slug>.md pointer (type: repos, source: → ../../../<book_dir>/, tagged ingest-by-reference) plus a compiled wiki/topics/<slug>.md overview. An article, by contrast, is copied in full into raw/articles/. Skim large books (index + representative chapters); do not flatten.
• Collection-scale data goes in a datasets/ manifest, not per-item files. Hundreds-to-thousands of row-like records (the ~20,900-card Synergetics Dictionary) belong in a datasets/ manifest — wiki as index, data external — optionally with one corpus inventory record. Surface the boundary explicitly and let the user override (they chose 365 individual inventory files for the holdings, which is their call).

Content discipline

• Establish authorship from the work's own text, not its slug or folder. In a single-subject collection, a Fuller-looking slug does not mean Fuller wrote it: sunspots is Steve Baer, est is L. Clark Stevens, we-are-all-astronauts is an exhibition catalog, the Coxeter geometry titles are third-party. Tag accordingly — buckminster-fuller (by) / about-buckminster-fuller / fuller-adjacent.

See Also

• Legacy Raw Source Registry — where the repo's external source files live, the layer these tooling lessons operate on
• Lemon Wiki (LLM Wiki) (Lemon Wiki (LLM Wiki)) — the agent-compiled wiki system this article lives in

Sources

• 2026-06-27-ll-wiki-migration-tooling.md — the raw session lessons-learned note (7 lessons) this article synthesizes