| name | pm |
| description | Obsidian Workspace PM — tasks, documents, and ADRs in your Obsidian vault. Triggers on task/doc/ADR lifecycle requests ("add a task", "create an ADR", "archive X", "list in-progress tasks", "refresh dashboard") and via `/obw:pm`. Requires `.obsidian.yaml` with a `pm.project` section. |
pm — Obsidian Project Management
Run the obsidian CLI directly — no sub-agent. For CLI syntax defer to the obsidian:obsidian-cli skill; this skill owns only the PM conventions: folder layout, template names, property schema, ADR numbering, and dashboard generation.
CLI gotcha: file= resolves like a wikilink — bare note name only, no path, no .md. Use path= for vault-root-relative paths. If a command returns File not found after a create, fix the parameter — never re-run create (it makes name 1.md duplicates).
Config (.obsidian.yaml)
vault: <VAULT_NAME>
pm:
project: <PROJECT_NAME>
Missing config or pm.project → tell the user to run /obw:init and stop. Never guess the vault.
Vault Layout
pm/
├── {project}/
│ ├── tasks/ # active tasks
│ ├── archive/ # completed tasks
│ └── docs/ # docs + ADRs
└── dashboard.base # cross-project dashboard (optional)
Folders are created on demand — create / move auto-create missing parent folders. No mkdir needed.
Template Names
Fixed: task, doc, adr. Installed into the vault's Obsidian Templates folder by /obw:init. Use via CLI create template=<name>.
If obsidian vault=<v> templates doesn't list one of these, /obw:init hasn't run (or the user removed them). Tell the user to re-run init rather than inlining template content here.
Operations
Use one call per known-name read — never chain search → read. The pm-specific bits:
- Create task →
create at pm/{project}/tasks/{name}.md with template=task, then set properties project / priority / due / tags.
- Create doc →
create at pm/{project}/docs/{name}.md with template=doc, then set project.
- Create ADR →
create at pm/{project}/docs/adr-{NNNN}-{title}.md with template=adr, then set project / status. See ADR numbering below.
- List tasks →
search with query="[type:task] [project:{project}] [status:<s>]" format=json.
- Archive → set
status=done and completed, then move to pm/{project}/archive.
- Delete → confirm first; fall back to
move if the build lacks delete.
ADR numbering
Before creating an ADR, search with query="[type:adr] [project:{project}]" format=json and take max(number)+1, zero-padded to 4 digits.
Property Schema
Task: type: task, status (todo/in-progress/blocked/done), priority (high/medium/low), project, due (date), tags (list), created, completed.
Doc: type: doc, project, created, updated.
ADR: type: adr, project, status (proposed/accepted/deprecated/superseded), created, deciders.
Property names are lowercase. Do not invent fields — dashboards depend on this schema.
Dashboards
Dashboards are Obsidian Bases (.base files — core in 1.9+). For Bases schema / filter / formula syntax, defer to the obsidian:obsidian-bases skill.
Generated from plugin templates via shell (template contents never enter context):
- Cross-project →
pm/dashboard.base:
obsidian vault={vault} create path="pm/dashboard.base" \
content="$(cat "${CLAUDE_PLUGIN_ROOT}/templates/dashboard-cross.base")" overwrite
- Per-project →
pm/{project}/dashboard.base:
obsidian vault={vault} create path="pm/{project}/dashboard.base" \
content="$(sed "s/__PROJECT__/{project}/g" "${CLAUDE_PLUGIN_ROOT}/templates/dashboard-project.base")" overwrite
Conversation-mode status (user asks in chat, not Obsidian): run the equivalent search and format a summary table in the reply. Don't write a .base file unless asked to.
Important Rules
- Read
.obsidian.yaml before any operation.
- Never
search to locate a note whose name is known — go straight to read.
- Never bypass the CLI with filesystem Read/Write against the vault. Only exception: reading
.obsidian/templates.json / obsidian.json during /obw:init. Dashboard creation uses the CLI via shell-piped content.
- Confirm destructive intents (delete, archive-move, ADR supersede) before executing.
- A claim of "created" / "updated" needs a receipt — the CLI's own success output counts; an error output never does. Report failures as failures.
- Bulk scans (e.g. auditing all archived tasks) may be delegated to a read-only Explore agent to keep the listing out of context; single-entity operations never need one.
