When building static sites, reusing a master template (like base.html) is standard practice. The layout defines the HTML head, script inclusions, navigation headers, and footer.

However, when you step outside the boundaries of your static site generator’s default compiler to build custom standalone pages (such as a public analytics dashboard or a Kanban board), you introduce a classic developer pitfall: the template scope leak.

Here is a look at how this bug manifests, why it happens, and how to resolve it at the build pipeline layer in Clojure/Babashka without pulling in bloated template engine libraries.

1. The Anatomy of the Leak

In our Babashka-native blog compiler (quickblog), standard markdown posts are parsed, and unused conditional tags (like a GitHub link or a Twitter handle) are cleanly stripped out.

But when rendering our custom /stats.html and /kanban.html dashboards, we bypassed the markdown parser and did a raw find-and-replace on the base template:

;; Bypassing the compiler layout context:
(clojure.string/replace base-html "{{body | safe }}" stats-body)

Because this operation was a simple string substitution, none of the template compiler’s conditional logic was evaluated. The browser received raw, unparsed control directives directly in the markup:

{% if discuss-link %}<a href="{{discuss-link}}">Discuss</a>{% endif %}
{% if twitter-handle %}<a href="https://twitter.com/{{twitter-handle}}">Twitter</a>{% endif %}

The browser ignored the unknown template syntax wrapper and rendered the text inside. Suddenly, disabled links (“Discuss”, “Twitter”, “About”) leaked onto the live layout as raw, broken navigation menu items.

2. The Heavy Fix vs. The Pragmatic Fix

To resolve this, you have two architectural choices:

Approach A: Pull in a Templating Library (Bloat)

You could add a full template compilation dependency (like Selmer or stencil) to your build runner. While robust, this adds compile-time dependency overhead, slows down local file-watchers, and introduces library complexity just to render two pages.

Approach B: Clean the Output at the Pipeline Layer (Zero-Dependency)

Since the master compiler already strips these values for standard pages, we can write a simple, stateless regex post-processing filter directly inside our Babashka task script (bb.edn):

(let [clean-template (fn [s]
                       (-> s
                           ;; 1. Strip out unparsed JTE control blocks
                           (clojure.string/replace (re-pattern "\\{%.*?%\\}") "")
                           ;; 2. Strip out remaining unused placeholder vars
                           (clojure.string/replace (re-pattern "\\{\\{favicon-tags \\| safe\\}\\}") "")
                           (clojure.string/replace (re-pattern "\\{\\{sharing\\..*?\\}\\}") "")
                           ;; 3. Remove conditional links that leaked
                           (clojure.string/replace (re-pattern "<a class=\"page-link\" href=\"\\{\\{discuss-link\\}\\}\">Discuss</a>") "")
                           (clojure.string/replace (re-pattern "<a class=\"page-link\" href=\"https://twitter.com/\\{\\{twitter-handle\\}\\}\">\\s*Twitter\\s*</a>") "")
                           (clojure.string/replace (re-pattern "<a class=\"page-link\" href=\"\\{\\{about-link\\}\\}\">About</a>") "")
                           ;; 4. Normalize page suffix
                           (clojure.string/replace (re-pattern "\\{\\{page-suffix\\}\\}") ".html")))]
  ;; Apply compiler filter to the rendered output:
  (spit "public/kanban.html" (clean-template raw-rendered)))

3. Resume Perspective: Why It Matters

This debugging exercise demonstrates three key engineering principles:

  1. Understanding Compiler Stages: Recognizing that string-level replacement does not inherit template-level parsing rules.
  2. Avoiding Dependency Inflation: Resisting the urge to add packages to solve a basic text-parsing problem. A few lines of targeted Clojure string replacements maintain a lightweight, fast, and zero-maintenance build pipeline.
  3. Defense-in-Depth UI: Ensuring client-facing pages never render unparsed server-side variables or control directives.