Scans the knowledge directory, finds units not yet in any index, and adds them. Running this skill repeatedly is safe and always converges — it only ever adds missing entries, never rewrites existing ones.
\nDomain index files live at <knowledge-dir>/_indexes/<Domain>.md. The compile skill matches a note to a domain by the index filename alone (e.g. AI.md, Business.md).
A unit may appear in at most 2 indexes: one primary domain, one secondary only when the unit genuinely straddles two.
\nRun all steps to completion without stopping. Only speak at the end (Step 4 summary).
\n_knowledge/ (relative to cwd, which is always the vault root)The index directory is always <knowledge-dir>/_indexes/. Create it if it does not exist.
Scan all knowledge units:
\nfind <knowledge-dir> -name \"*.md\" -not -path \"*/_indexes/*\" | sort\nScan all existing index files and collect every Path value from their tables:
find <index-dir> -name \"*.md\" | sort\nA unit is unindexed if its vault-absolute path (e.g. /_knowledge/Vertical Agent.md) does not appear in any index table. Build a list of only these units.
If all units are already indexed, report and stop.
\nFor each unindexed unit, read its YAML front matter and first non-heading paragraph. Build an internal list:
\nreferences/extract.md documents partial-read commands — vaultr extract tag retrieves front-matter tags directly; vaultr extract segment --head N retrieves the opening lines. Use them instead of a full vaultr read when only these fragments are needed.
title | entity_type | tags | summary (≤80 chars) | vault-absolute path\nVertical Agent | concept | [ai, agent] | AI product focused on a specific vertical domain | /_knowledge/Vertical Agent.md\nJane Smith | person | [founder, ai] | Founder of SophiaPro, AI startup | /_knowledge/Jane Smith.md\nFor each unindexed unit, determine domain assignment:
\nMatch existing domains first by semantic understanding of the domain name (from existing index filenames) against the unit's title, entity_type, tags, and summary. Assign the strongest match.
\nSecondary domain only when the unit genuinely belongs to two domains — passing relevance is not enough. When in doubt, assign one.
\nNew domain when no existing domain fits. Create a new domain if the topic is broad and stable enough to plausibly accumulate more units over time (e.g. Physics, History, Health, Law). The test is domain breadth, not how many unindexed units are present right now — a single unit can justify a new domain. Only fall back to \"General\" when the topic is genuinely narrow or one-off and unlikely to grow.
Domain names should be broad and stable (e.g. AI, Business, Science, People, Philosophy). Don't name a domain after a single entity unless it anchors a large cluster.
For each affected domain index, append the new rows and update unit_count and updated in the frontmatter. If the index file does not exist yet, create it:
---\nkind: index\ndomain: <Domain>\nunit_count: <N>\nupdated: <YYYY-MM-DD>\n---\n\n# <Domain>\n\n| Title | Type | Tags | Summary | Path |\n| -------------- | ------- | ------------- | ------------------------------------------------ | ----------------------------- |\n| Vertical Agent | concept | [ai, agent] | AI product focused on a specific vertical domain | /_knowledge/Vertical Agent.md |\n| Jane Smith | person | [founder, ai] | Founder of SophiaPro, AI startup | /_knowledge/Jane Smith.md |\nKeep rows sorted alphabetically by Title within each index file.
\n