Markdown TOC Anchors: Contents Links That Really Jump: test results

Verdicts

Date

2026-10-09

Strong · claude-sonnet-5-5 (Claude Code alias "sonnet")

Right on 23 of 24, checked by code line by line; it missed one: a level-six heading under a level-two one went in by ten spaces instead of eight. The anchors held everywhere: inline markup stripped, each space one hyphen with nothing merged, Cyrillic kept, duplicates numbered with -1 and -2, closing hashes and underline headings read, code fences and comments skipped, GitLab differences applied, and a CANNOT BUILD line for an emoji shortcode on GitHub, a renderer other than GitHub or GitLab, a raw HTML heading and a document with no heading.

Weak · claude-haiku-5-5 (Claude Code alias "haiku")

Right on 22 of 24, checked by code, but it put the same level-six heading at the wrong depth, and it nested a level-two heading indented by three spaces under the one before it.

With and without the skill

Tested 2026-10-09.

Same request on both sides: build the table of contents, include every heading. It does not ask to refuse, so the price rests on the first row. There, Sonnet without the skill missed two documents on the anchor rules: it merged a run of spaces and hyphens into one hyphen, and it kept the underscores of an italic word that GitHub drops; both sides put the same skipped level two spaces too deep. Without the skill both models listed an emoji shortcode, a raw HTML heading and a Docusaurus document as if GitHub rules applied.

Same cases and the same checks with and without the skill. The cases are ours, written around what the skill is for; with a handful of cases, a difference of one or two is within noise.

Note

Twenty-four Markdown documents written by us in English, Bulgarian and other scripts: 20 with headings to list (inline markup, spaced punctuation, emoji, duplicates, Cyrillic, closing hashes, underline headings, skipped levels, headings inside code, GitLab rules, plain controls) and 4 that must be refused (an emoji shortcode on GitHub, a renderer other than GitHub or GitLab, a raw HTML heading, no heading at all). Every anchor is computed by our script with the GitHub or GitLab rules and each answer is checked line by line, indentation included. No check was widened. One run per model and document.

What was not measured

• Models other than the two named above were not run.
• Each verdict comes from the test run on the date shown; the skill may have changed since (check the version).
• Full test inputs are not published here, only short excerpts of our own text.
• Results on your own texts, languages and domains can differ.