# Obsidian Wiki Doctor > Use this whenever the user wants to diagnose, validate, score, report on, or explain the health or maturity of an Obsidian LLM Wiki, including prompts like "run Obsidian Wiki Doctor", "诊断 wiki", "给 wiki 打健康分", "出中文健康报告", "看看初始化后有没有用", or questions about whether poor query quality is caused by wiki structure. This skill is read-only and never repairs files. - Skill: `huajiexiewenfeng/obsidian-wiki-doctor` (Agent Skill, multi-file: 4 files) - Install (CLI): `npx skillmds@latest add huajiexiewenfeng/obsidian-wiki-doctor` - Raw SKILL.md: https://api.skillmd.com/api/skills/huajiexiewenfeng/obsidian-wiki-doctor/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: AI & ML - Author: huajiexiewenfeng (https://skillmd.com/u/huajiexiewenfeng) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/huajiexiewenfeng/obsidian-wiki-doctor --- ## First-Use Vault Setup Treat root configuration as background setup, not a JSON-file task for the user. If normal resolution has no root, run `python "" root discover --format json`. Show returned existing absolute paths as numbered candidates and ask the user to select one or provide another absolute Vault path. Resolve the selected path and state `vault_root`, `control_center`, and `wiki_root`. Only after the user confirms it should become the default, run `root configure --root --activate --confirm`. Do not read note content or scan the whole disk during discovery. Continue the user's original request after setup succeeds. ## Runtime Resolution Before running any command, resolve this skill's `SKILL.md` directory, take its parent as ``, and set: ```text = /obsidian-wiki-runtime/scripts/llm_wiki.py ``` Verify that `` exists, then invoke it by absolute path. If it is missing, stop with `missing-runtime`, report the expected path, and recommend: ```text npx skills add huajiexiewenfeng/obsidian-llm-wiki --skill '*' --copy --yes ``` Do not fall back to a repository-relative `scripts/llm_wiki.py` path. # Obsidian Wiki Doctor ## Inventory Discovery Doctor calls the shared Inventory Core on every resolved Vault and remains strictly read-only. It compares supported document metadata with `.meta/inventory.json`, `.meta/sources.json`, and `.meta/pages.json`. - `missing-ingest-inventory`: show an initialize dry-run; never initialize automatically. - Initial baseline documents reachable from `wiki/index.md` are `known-existing`; disconnected historical documents are `unverified`, not new ingest candidates. - `source-island`: an `unverified` historical document is outside the rooted graph. - `source-coverage-lost`: a `known-existing` document lost its rooted graph connection. - `uningested-source`: a `discovered` post-baseline document lacks processed evidence. - `stale-ingested-source`: processed source metadata changed after ingest. - `orphan-wiki-page`: a Wiki page is not reachable from `wiki/index.md`. - `detached-wiki-component`: two or more Wiki pages form a disconnected component. - Invalid baselines, collisions, or incomplete scans block completeness claims. - Sensitive scopes expose alias/count summaries only. ### Mandatory Missing-Baseline Explanation Whenever `missing-ingest-inventory` is present, the Chinese report must state these three lines verbatim. Do not merely say that all current documents need classification or tracking: ```text 索引可达的历史文档 -> known-existing -> 不需要 ingest 不可达的历史文档 -> unverified/source-island -> 不自动 ingest 基线后新增 -> discovered/uningested-source -> 待处理 ``` Explain that the initial baseline classifies existing history; it does not turn the current Vault into an ingest backlog. When required `.meta` state files are also missing, recommend state initialization first, then Inventory initialize dry-run and explicit confirmation. Never imply that the current document count equals the number of documents requiring ingest. Use `inventory inspect --format json` for the complete non-sensitive candidate list. Text output groups large sets and shows at most 20 examples per directory. Graph traversal reads only ordinary Inventory documents and Wiki pages; excluded and sensitive scopes are never opened for relationship discovery. Diagnose an Obsidian LLM Wiki with the deterministic doctor engine. ## Boundary Use this skill to look, score, validate, and explain. Do not edit vault files. Use `obsidian-wiki-maintain` when the user asks to fix, repair, patch links, update `index.md`, add source proxies, or apply findings from a doctor report. ## Wiki Root Resolution Before reading, resolve and state `vault_root`, `control_center`, and `wiki_root` using the shared order: user path, nearest project `.obsidian-llm-wiki.json`, `OBSIDIAN_LLM_WIKI_ROOT`, then exactly one active Vault in user configuration. Otherwise stop with `missing-config` or ask the user to choose when multiple roots exist. Do not search the whole disk. ```text python "" root resolve --cwd --format json ``` ## Commands Human report: ```text python "" doctor report --root --format text ``` Machine validation: ```text python "" doctor validate --root --format json --fail-on error ``` Structured score: ```text python "" doctor score --root --format json ``` `/obsidian-wiki-runtime/scripts/obsidian_wiki_doctor.py` remains compatible in v0.2. ## Interpretation Rules - Treat script findings as deterministic evidence. - Treat score as directional guidance, not a KPI. - Never print secret values. - Keep Chinese-first explanations for Chinese users. - Explain `not-applicable` dimensions instead of treating them as failures. - If a sensitive-pattern finding exists, recommend verification first. Only a confirmed real credential should be rotated and prioritized ahead of state initialization; a suspected or redacted example does not block Inventory. - If the user asks to repair, hand off to `obsidian-wiki-maintain` with a narrow repair scope. - Explain archive registry/path/checksum drift, unregistered `raw/` files, and orphan archive staging files as read-only findings; never repair them here. ## References - Read `references/doctor-checks.md` when explaining finding names. - Read `references/report-template.md` when summarizing report structure. - Read `references/safety-rules.md` before presenting sensitive-pattern findings.