Zettelclaw

Zettelclaw

Follow the canonical Zettelclaw vault model: capture externally, write durable notes manually, and use the agent for navigation/synthesis.

This is an instruction-only skill. It does not install software by itself. qmd is optional when it is already available in the environment, and rg is the supported fallback for search.

Vault structure

<vault>/
├── 00 Inbox/
├── 01 Notes/
├── 02 Journal/
├── 03 Templates/
├── 04 Attachments/
└── README.md

Note types

Use YAML frontmatter on every note:

• Required on all notes:
  • type
  • tags
  • created
• Required on doc and content notes:
  • status (queued | in-progress | done | archived)
• Optional content metadata:
  • author
  • source

Use these primary types:

• note: durable atomic thinking note; no status
• doc: non-atomic working/reference note; uses status
• journal: daily log note; no status
• content types: article, book, movie, tv, youtube, tweet, podcast, paper (and extensible additional content types); uses status

Templates

Always read the matching template in 03 Templates/ before creating a note:

• note.md
• journal.md
• clipper-capture.json

Use core templates/date syntax. Do not require Templater.

Inbox workflow

• Web captures land in 00 Inbox/ via clipper-capture.json.
• Clipper sets type by URL (tweet, youtube, else article) and status: queued.
• Process inbox captures by keeping/moving, converting into a type: note, or deleting.
• Do not auto-write durable thinking notes from captured content unless explicitly asked.

Bases workflow

• 00 Inbox/inbox.base is the canonical queue view.
• Grouping is by note.type for scan-by-content-type triage.
• Prefer creating/editing .base files over Dataview.

Titles as APIs

A note's title is its interface. Use complete, declarative phrases: "Spaced Repetition Works Because of Retrieval," not "Spaced Repetition." A well-titled note can be linked and understood without opening it. When creating or renaming notes, always prefer a full declarative statement.

Tag conventions

Tags live in the frontmatter tags array, not inline in the body.

• Lowercase, hyphenated: spaced-repetition, not Spaced Repetition or spacedRepetition.
• Topic-oriented, not structural: learning (what it's about), not important (how you feel about it).
• Nest only when a hierarchy is genuinely useful: ai/transformers is fine, deep nesting is not.
• Suggest tags based on the vault's existing taxonomy rather than inventing new ones.

Editing rules

• Preserve existing prose unless user asks to rewrite.
• Do not add or maintain an updated frontmatter field.
• Use dense wikilinking ([[Note Title]]) and allow unresolved links as stubs.
• Do not create top-level folders unless explicitly requested.
• Do not assign/change status, move notes, or delete notes without explicit instruction.
• Agent write surface is limited to:
  • /ask response callouts
  • optional daily briefing callout in journals

Search patterns

Default QMD collections:

• zettelclaw-inbox
• zettelclaw-notes
• zettelclaw-journal
• zettelclaw-attachments

# qmd (preferred when installed)
qmd query "spaced repetition and retrieval" -c zettelclaw-notes
qmd search "status: queued" -c zettelclaw-inbox
qmd vsearch "what have I been learning about memory" -c zettelclaw-notes

# ripgrep fallback
rg -l 'type: note' "01 Notes/"
rg -l 'type: article' "00 Inbox/" "01 Notes/"
rg -l 'status: queued' "00 Inbox/" "01 Notes/"

OpenClaw integration

If configuring OpenClaw memory paths, use:

• agents.defaults.memorySearch.extraPaths

Do not write legacy top-level memorySearch. Only change OpenClaw config when the operator explicitly wants this vault wired into an OpenClaw installation.