Wiki setup

Wiki setup, interview schema and substitution rules

Standard entry-point suggestions

When the user runs /setup-wiki, offer this catalog. The user picks which to enable and may override paths. Defaults are conservative; nothing is enabled unless the user opts in.

Suggestion key	Default path	source_type	Default quality	Default post_ingest	Naming convention
quick-notes	99_Quick-notes/	quick-note	0.5	move	YYYY-MM-DD Short title.ext
other	98_Other/	article	0.6	move	YYYY-MM-DD <slug>.ext
voice-transcripts	2_Transcripts/	voice-transcript	0.5	keep	(vendor-defined; no rename)
daily-journal	1_Journal/	daily-note	0.45	keep	YYYY-MM-DD.md
llm-conversations	4_Claude-conversations/	claude-chat	0.3	move	YYYY-MM-DD <slug>.md
inbox-pdfs	97_PDFs/	article	0.6	move	YYYY-MM-DD <slug>.pdf
images	96_Images/	image	0.4	move	YYYY-MM-DD <slug>.<ext>

For each enabled suggestion, ask:

1. Confirm or override the path.
2. Confirm or override default_quality.
3. Confirm or override post_ingest (move, keep, or read_only).
4. Confirm or override the naming convention.
5. Optionally specify an exclude list (filenames to skip during ingest).

The user may also add custom entry points beyond this catalog. For each custom one, ask path, source_type (must be a bucket listed in wiki-core SKILL.md), default_quality, post_ingest, naming_convention, exclude.

post_ingest semantics

Value	What happens to source files after ingest
move	Add processed: true and processed_at: YYYY-MM-DD frontmatter, then relocate the file to _service/entry-points/<entry-point-name>/<YYYY-MM>/.
keep	Add processed: true and processed_at: YYYY-MM-DD frontmatter. Leave the file in place. Monthly archival (if applicable) still moves files dated before the current month into <entry-point>/YYYY/YYYY-MM/ subfolders.
read_only	Do NOT add any frontmatter. Do NOT move the file. The source is treated as immutable. Manifest records the SHA-256 to dedupe across runs. Monthly archival still moves files dated before the current month into the date subfolders.

read_only is the strictest setting; use it for entry points whose contents must never be modified by the agent (e.g. archive folders the user maintains by hand, vendor-managed folders, or sync source-of-truth folders).

exclude list per entry point

- path: "1_Journal/"
  source_type: daily-note
  default_quality: 0.45
  post_ingest: keep
  naming_convention: "YYYY-MM-DD.md"
  exclude:
    - "_template.md"
    - "_draft-*.md"

Glob patterns are matched against the filename relative to the entry-point root. Excluded files are never processed, never moved, never have frontmatter added. They also are not flagged as orphans by /lint.

Standard structured-knowledge suggestions

Suggestion key	Default path	Purpose	Routing hint
projects	1_Projects/	projects	Active work, organized by intent category
documentation	3_Documentation/	documentation	LLM-curated knowledge articles
resources	2_Resources/	resources	Lists, references, recipes, places, contacts
people	5_People/	people	Person pages
concepts	6_Concepts/	concepts	Companies, markets, frameworks, technical concepts

The user picks which to enable, overrides paths, and may add custom folders. Each folder gets a stub <folder-name>.md index after scaffolding.

Interview answer schema

The interview produces this object, which the setup command persists into CLAUDE.md and the registry:

name: "Wiki display name"
slug: "wiki-slug"
root: "/absolute/path"
entry_points:
  - path: "99_Quick-notes/"
    source_type: quick-note
    default_quality: 0.5
    post_ingest: move
    naming_convention: "YYYY-MM-DD Short title.ext"
structured_knowledge:
  - path: "1_Projects/"
    purpose: projects
    routing_hint: "Active work, organized by intent category"
dashboards:
  - path: "0_To-do.md"
    type: todo
  - path: "0_Board.md"
    type: dataviewjs-board
protected_paths:
  - "2_Resources/Recipes/"
tags:
  - "#work"
  - "#research"
writing_style: "No empty line after headings. No trailing summaries unless asked. Inline provenance markers on documentation pages only."
project_thresholds:
  active_to_dormant_months: 6
  dormant_to_archive_months: 4
  completed_to_archive_months: 6
templates_to_install:
  - todo-dashboard
  - daily-note
  - canvas-dashboard

Per-wiki files written at setup

Setup writes two files at the wiki root and several inside _service/:

File	Source template	Substitution?
CLAUDE.md	templates/CLAUDE.md.tmpl	No, copied verbatim
wiki-config.md	templates/wiki-config.md.tmpl	Yes, all placeholders below
index.md	templates/index.md.tmpl	{{wiki_name}} only
_service/.manifest.json	templates/manifest.json.tmpl	{{iso_timestamp}}
_service/log.md	empty	—
_service/hot.md	templates/hot.md.tmpl	{{iso_timestamp}}
_service/feedback.md	templates/feedback.md.tmpl	{{today}}

CLAUDE.md is intentionally identical across every wiki this plugin manages. It must not contain any wiki-specific data. All wiki-specific data lives in wiki-config.md.

Template substitution map (for wiki-config.md and the dashboard templates)

Placeholder	Source
{{wiki_name}}	name
{{wiki_slug}}	slug
{{wiki_root}}	root
{{wiki_root_basename}}	basename of root (used in Tasks/Dataview queries)
{{projects_path}}	first structured_knowledge entry with purpose: projects; empty if none
{{documentation_path}}	first entry with purpose: documentation
{{resources_path}}	first entry with purpose: resources
{{people_path}}	first entry with purpose: people
{{concepts_path}}	first entry with purpose: concepts
{{quicknotes_path}}	first entry point with source_type: quick-note
{{other_path}}	first entry point with source_type: article
{{journal_path}}	first entry point with source_type: daily-note
{{transcripts_path}}	first entry point with source_type: voice-transcript
{{conversations_path}}	first entry point with source_type: claude-chat
{{tags_block}}	YAML-rendered list, each tag indented two spaces with - prefix
{{writing_style_indented}}	writing_style text, indented two spaces for YAML pipe block
{{active_to_dormant_months}}	project_thresholds.active_to_dormant_months
{{dormant_to_archive_months}}	project_thresholds.dormant_to_archive_months
{{completed_to_archive_months}}	project_thresholds.completed_to_archive_months
{{entry_points_block}}	YAML-rendered entry_points array (- items, two-space indent). Includes exclude when set.
{{structured_knowledge_block}}	YAML-rendered structured_knowledge array
{{dashboards_block}}	YAML-rendered dashboards array (empty list [] if none). Each entry carries path and a free-form type label (todo, canvas, dataviewjs-board); the label is not validated
{{protected_paths_block}}	YAML-rendered protected_paths array (empty list [] if none)
{{ignore_paths_block}}	YAML-rendered ignore_paths array (empty list [] if none)
{{pii_paths_block}}	YAML-rendered pii_paths array (empty list [] if none)
{{projects_path_quoted}}	<wiki_root_basename>/<projects_path>, trailing slash stripped (Dataview FROM clause)
{{transcripts_path_quoted}}	<wiki_root_basename>/<transcripts_path>, trailing slash stripped (Dataview FROM clause)
{{transcripts_exclude_path}}	<wiki_root_basename>/<transcripts_path> (Tasks path does not include filter)
{{conversations_exclude_path}}	<wiki_root_basename>/<conversations_path> (Tasks path does not include filter)
{{today}}	current date YYYY-MM-DD
{{iso_timestamp}}	current timestamp ISO-8601

Placeholders left unresolved by this map are an error: abort and report which placeholder failed.

Dashboard template conditional rules

The canvas dashboard template references entry points and folders the wiki may not have configured. When substituting dashboard.canvas.tmpl (and only dashboard templates):

• If {{transcripts_exclude_path}} or {{conversations_exclude_path}} resolves empty (the corresponding entry point is not configured), delete the entire path does not include … line from each query rather than leaving an empty filter.
• If {{projects_path_quoted}} resolves empty (no purpose: projects folder), remove the whole projects_table node from the canvas JSON and tell the user it was skipped.
• If {{transcripts_path_quoted}} resolves empty (no voice-transcript entry point), remove the whole recent_transcripts node and tell the user it was skipped.

These removals are not errors; report them in the setup summary.

Validation rules before writing to disk

Refuse to scaffold if any of these fail:

• slug is not kebab-case or short (lowercase letters, digits, hyphens; cannot start with a digit; max 32 characters).
• slug already exists in the registry and the user is in new-wiki mode.
• root is not absolute.
• root is inside another registered wiki's root (no nested wikis).
• Any entry_points[].path or structured_knowledge[].path overlaps with another (path prefix collision within the wiki).
• Any source_type is not listed in wiki-core SKILL.md "Source quality buckets".
• Any post_ingest value is not one of move, keep, read_only.
• project_thresholds.* are not positive integers.
• Any custom_procedures[].when is not one of pre-ingest, during-ingest, post-ingest, pre-lint, post-lint.
• Any custom_procedures[].procedure resolves to a path outside the wiki root.

After scaffolding succeeds, the agent creates an empty <wiki-root>/_service/custom-procedures/ folder for any custom procedures the user declared, and copies templates/_custom-procedure.md.tmpl to each declared procedure path so the user has a starter file to edit.

Registry write protocol

Read ~/.claude/obsidian-wiki/wiki-registry.json once. Compute the new state. Write to ~/.claude/obsidian-wiki/wiki-registry.json.tmp. Verify the JSON parses. Rename atomically to wiki-registry.json.

If the registry file is malformed (does not parse), abort and ask the user to fix it before proceeding. Never overwrite a malformed registry; that would lose all other wiki entries.

Schema:

{
  "version": 1,
  "vault_root": "/absolute/path/to/vault",
  "wikis": {
    "<slug>": {
      "name": "Display Name",
      "root": "/absolute/path/to/wiki/root",
      "created": "ISO-8601"
    }
  }
}

vault_root is established at first wiki registration (default: parent of the first wiki's root, asked of the user once). All wikis share it. The shared docs live at <vault_root>/_service/docs/.

Shared docs install

Every /setup-wiki run, after the registry is updated, refreshes <vault_root>/_service/docs/:

1. Create the directory if missing.
2. Copy ${CLAUDE_PLUGIN_ROOT}/README.md to <vault_root>/_service/docs/README.md. Overwrite.
3. Copy ${CLAUDE_PLUGIN_ROOT}/docs/diagrams/*.svg to <vault_root>/_service/docs/diagrams/. Overwrite each.

The same refresh logic is exposed as /update-docs for between-setup refreshes (e.g. after the plugin is updated via /plugin update obsidian-wiki).

Optional: session-end capture hook

After setup succeeds, offer to install the stop hook that prompts /capture --quick at the end of any Claude Code session that had file edits or significant shell activity:

1. Ask: "Install the session-end capture hook? It will nudge you to run /capture --quick after productive sessions." [y/n]
2. If yes, copy ${CLAUDE_PLUGIN_ROOT}/.claude/hooks/wiki-stop-capture.sh to ~/.claude/hooks/wiki-stop-capture.sh (create dir if missing).
3. Read ~/.claude/settings.json (create as {} if missing). Add the hook entry:

   {
     "hooks": {
       "Stop": [
         {
           "hooks": [
             {
               "type": "command",
               "command": "bash ~/.claude/hooks/wiki-stop-capture.sh"
             }
           ]
         }
       ]
     }
   }
   

   Merge with existing content; do not overwrite other hooks.
4. Confirm: "Stop hook installed globally. It activates in every Claude Code project."

The hook fires at most once per session (session_id sentinel in TMPDIR) and never fires if the turn was itself injected by a hook (stop_hook_active guard).

Removing a wiki

Reconfigure mode supports a --remove flag. When set:

• Confirm twice.
• Delete the registry entry.
• Do NOT delete the wiki's folder or any of its content. The user removes their data manually.