메뉴
Read-only vault health check. Use when user says 'how's the vault', 'audit', 'lint', 'anything broken', 'show me broken links/stubs/orphans'. Do NOT use when user wants fixes applied — hand off to autoimprove.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Answer questions from vault notes. Use when user asks 'how does X relate to Y', 'what do the papers say about Z', 'compare approaches to W', or discusses concept connections. Also fire immediately after wiki-ingest when user pivots from saving to asking how the new source relates to other vault topics. Also fire as a pre-ingest relevance check: before newly-shared content is dismissed as not vault-worthy, check whether it connects to existing vault notes. Also fire proactively as a conversation develops: when discussion drifts near vault topics, surface a genuinely non-obvious connection the vault holds — sparingly, only when it clears the bar. Do NOT answer from memory — read the actual notes.
ALWAYS use when the user wants a source saved to the vault — pastes an arXiv ID/URL, says 'add this to the vault', 'save this', 'remember this', 'I found this interesting', or describes a specific finding/technique/project to capture. Do NOT activate for general topic conversation — only when there's explicit save intent or a concrete source (URL, paper ID, pasted content).
Autonomous vault improvement loop (writes changes). Use when user says 'improve the vault', 'autoimprove', 'fix what you can', 'what's the vault score'. Do NOT use for read-only diagnostics — that's wiki-lint.
Fill concept stubs with citation-grounded definitions. Use when user says 'fill in the stubs', 'compile concepts', 'flesh out concepts', or as follow-up after wiki-lint reports stubs. Do NOT use for new sources not yet in vault — that's wiki-ingest.
User-invoked: surface hidden concept connections via local embeddings (Ollama + nomic-embed-text). Flags related-but-unlinked (sentence, concept) pairs, then adds wikilinks where the connection holds. Invoke explicitly via /wiki-deep-link.
Set up, list, rename, or merge research domains. Use when user says 'add a domain for X', 'set up a domain', 'what domains exist', 'change scope of X domain', 'merge X and Y domains'. Do NOT speculatively fire on ingest prompts — wiki-ingest hands off explicitly when no domain matches.
| name | wiki-lint |
| description | Read-only vault health check. Use when user says 'how's the vault', 'audit', 'lint', 'anything broken', 'show me broken links/stubs/orphans'. Do NOT use when user wants fixes applied — hand off to autoimprove. |
Run a comprehensive health audit on the Obsidian vault and automatically fix mechanical issues.
The vault accumulates entropy over time — broken links from renamed notes, stale MOC counts, stub concepts that never got filled in, malformed date artifacts from earlier tooling. This skill surfaces all of those issues in one report and fixes what it can mechanically, so the user only needs to deal with issues that require human judgment.
Run commonplace vault-path to get the vault path. Use it in all commands below.
Indexes may be stale. Run a full rebuild to ensure accurate lint results:
commonplace index
commonplace lint
This prints a human-readable summary with counts by severity:
Do not pipe to python3 or jq. The default output is already designed for you to read. If you genuinely need structured data (e.g. to drive a multi-step fix), write it to a file with commonplace lint --json > /tmp/lint.json and Read the file — never parse via shell one-liner.
Show a concise summary to the user. Example:
Vault Health Report:
- 23 critical issues (12 broken links, 8 malformed dates, 3 frontmatter errors)
- 57 improvements (52 stubs, 3 stale MOCs, 2 near-duplicate concepts)
- 14 suggestions (12 orphans, 2 malformed names)
- 33 issues are auto-fixable
Dispatch the wiki-linter agent to fix everything that's mechanical:
P25-11-07 artifacts)Agents have isolated context windows — they cannot see this conversation. Include the vault path and lint results inline in the agent prompt so it knows exactly what to fix.
After fixes:
After the mechanical report, surface actionable intelligence from the vault's shape:
High-value stubs — concept stubs with high backlinkCount are the most-referenced unknown concepts in the corpus. For the top 3-5:
Grep "isStub.*true" in concept-index.jsonl (path set to $VAULT_PATH/.wiki/concept-index.jsonl)
Sort by backlinkCount descending. For each, suggest: "[[ConceptName]] is referenced N times but has no definition. Suggested search: ..."
Bridge concepts — concepts appearing in 2+ domains (domains array length > 1) are cross-domain connectors worth deepening. Mention the top 2-3 and which domains they bridge.
Orphaned clusters — if orphan notes exist, check whether they share concepts with non-orphan notes. If so, they're disconnected from the graph but not truly isolated — suggest adding wikilinks to reconnect them.
Semantic contradictions (when there's time/budget) — read the top 5-10 concept notes by backlinkCount and check for conflicting definitions or claims that newer sources in the vault may supersede. Flag any contradictions for the user to resolve. This is a main-model operation, so only do it on explicit request or when the user asks for a "deep lint."
Log: append to $VAULT_PATH/.wiki/log.md:
commonplace log --entry "## [$(date +%Y-%m-%d)] lint | Full audit\n- Critical: N, Improvements: N, Suggestions: N. Fixed: N mechanical.\n"
The user can ask about specific checks:
commonplace lint --check stubs
Valid check names: unresolved, stubs, orphans, frontmatter, moc-staleness, scope-violations, duplicates, malformed-dates, near-duplicate-names, malformed-concept-names