knap-skill

A Claude Code skill for Knap, Kepano's template language that turns data into Markdown.

Not written from the docs. Every command in the skill was run against the Knap CLI (0.4.0), and it ships the gotchas that cost failed renders the first time round.

Install

git clone https://github.com/remigius-labs/knap-skill
cp -r knap-skill/knap ~/.claude/skills/knap

Claude Code picks it up on the next session. Knap itself: npm i -g knap (Node 20+).

What it covers

• render one note from a JSON object, batch one file per CSV row / JSON array item / JSON file in a folder
• Filename templates, --dry-run, --overwrite, stdin piping, Defuddle → Knap for URL-to-note
• Obsidian frontmatter, tags, wikilinks, callouts, and tables from data
• Every standard filter name (knap/filters.md), since unknown filter names are hard errors

The gotchas

Symptom	Fix
UNKNOWN_FILTER: "default"	Use {{ x ?? "fallback" }}
tags: - "a" broken YAML	{{ tags | list }} on its own line under tags:
["[[A]]","[[B]]"] in output	wikilink on an array returns JSON, add | join:", "
PARSE_ERROR: Unexpected character '-'	{{- / -}} trim is not supported in the CLI
Stray blank lines around loops	Put each {% %} tag on its own line, Knap removes that line
Time shifted by hours	date renders in local timezone, feed date-only strings

Example

---
type: lead
date: {{ date | date:"YYYY-MM-DD" }}
status: {{ status | lower }}
tags:
{{ tags | list }}
---
# {{ name }}

Channel: {{ channel ?? "unknown" }}
{% for n in notes %}
- {{ n }}
{% endfor %}
Related: {{ people | wikilink | join:", " }}

knap batch lead.md --data leads.json --output-dir notes \
  --filename '{{ date | date:"YYYY-MM-DD" }}-{{ name | safe_name | kebab }}.md' --dry-run

Output, notes/2026-09-10-ada.md:

---
type: lead
date: 2026-09-10
status: open
tags:
- lead
- friend
---
# Ada

Channel: unknown
- First call done
- Sent the deck
Related: [[Ada]], [[Lovelace Labs]]

How it was built

Test-driven: a fresh agent with no skill could not produce a single Knap command. The skill was written from real runs, then a fresh agent with the skill rendered a batch correctly first try.

MIT.