Templates

Filenames, file contents, and the commit message are all Nunjucks templates rendered once per submission. commit.prefix is the one exception: it is a literal marker and is not templated.

Variables

submission

Variable	Description
submission.id	LeetCode submission id (number)
submission.timestamp	Unix seconds
submission.date	YYYY-MM-DD (UTC)
submission.lang	Language id (python3, typescript, …)
submission.lang_ext	File extension mapped from the language (py, ts, …)
submission.code	The solution code
submission.status	e.g. Accepted
submission.runtime	e.g. 42 ms (null when unavailable)
submission.memory	e.g. 13.5 MB (null when unavailable)
submission.runtime_percentile	Number (null when unavailable)
submission.memory_percentile	Number (null when unavailable)

question

Variable	Description
question.frontend_id	e.g. 1
question.title	e.g. Two Sum
question.title_slug	e.g. two-sum
question.difficulty	Easy | Medium | Hard
question.tags	Array of topic tag names
question.url	https://leetcode.com/problems/<slug>/
question.content	Raw problem HTML (convert with toMarkdown / toTypst)
question.acceptance_rate	Number (null when unavailable)
question.is_paid_only	Boolean

Undefined variables render empty by default; set render.throwOnUndefined: true to make them throw instead — useful for catching typos.

Built-in filters

Every standard Nunjucks filter works: join, lower, replace, trim, truncate, capitalize, striptags, sort, groupby, dump, urlencode, and so on. leech adds a set of its own — see Filters & conversion.

A worked example

This produces a frontmatter-rich README, a Typst document, and the raw code:

files:
  - filename: "{{ question.title_slug }}/README.md"
    content: |
      ---
      title: "{{ question.title }}"
      id: {{ question.frontend_id }}
      difficulty: {{ question.difficulty }}
      date: "{{ submission.timestamp | datefmt('YYYY-MM-DD') }}"
      tags: [{% for t in question.tags %}"{{ t }}"{% if not loop.last %}, {% endif %}{% endfor %}]
      ---

      # {{ question.title }}

      {{ question.content | toMarkdown({ gfm: true }) }}

  - filename: "{{ question.title_slug }}.typ"
    content: |
      = {{ question.title }}

      {{ question.content | toTypst }}

      == Submission

      {{ submission.code | codeBlock(submission.lang) }}

  - filename: "{{ question.title_slug }}/{{ submission.lang }}.{{ submission.lang_ext }}"
    content: "{{ submission.code }}"

Filename sanitization

Rendered filenames are sanitized per path segment. Reserved characters (/\?%*:|"<> and control characters) and whitespace become -; runs of dashes collapse; leading and trailing dots and dashes are stripped; empty segments become untitled. This means a problem title with a slash or colon can't escape its directory or break on Windows checkouts.

Collisions and overwrites

Filenames are not unique by construction. If two submissions render the same path (for example, the same problem solved twice in the same language, or a README.md per problem), the later commit overwrites the earlier file. That's usually what you want — an improved solution replaces the old one — but if you need history preserved, include a unique key in the filename:

- filename: "{{ question.title_slug }}/{{ submission.id }}/{{ submission.lang }}.{{ submission.lang_ext }}"
  content: "{{ submission.code }}"

Pitfall: no nested braces inside filter arguments

A filter argument is a plain expression, so reference variables directly:

{{ submission.code | codeBlock(submission.lang) }}

Writing codeBlock({{ submission.lang }}) is a Nunjucks syntax error. It looks for a dict literal and fails with parseAggregate: expected colon after dict key, reported against the template's line and column — which is easy to mistake for a YAML error in the workflow file.