Skip to content
Agents tracked: 501 Releases (7d): 777 Agent status: no reported issuesSince last limit reset: Claude 10d · Codex 8d For AI agents: Markdown Updated Nov 21, 2026
Showcase Emil StenströmResults as reported by the author's blog

JustHTML: a Python HTML5 parser built with coding agents that passes html5lib tests

A developer spent his off-hours using Copilot agent mode with several models to build a dependency-free HTML5 parser, and wrote up every stage, including a full restart.

The problem

HTML5 parsing is notoriously complex. A large public conformance suite gave the agents a clear measure of progress, which made a mostly agent-written parser realistic.

How they did it

  1. Connect the html5lib-tests suite at the start and use its pass rate as the goal.
  2. Iterate with the agent, then restart from scratch by porting logic from html5ever.
  3. Profile, delete unused code found through coverage analysis, and harden the parser with fuzzing.

Results as reported by the author's blog

  • JustHTML is about 3,000 lines of Python with 8,500+ tests passing.
  • It passes 100% of the html5lib test suite.
  • It passed 3 million generated webpages without any crashes.

These figures are as reported by the author's blog in Friendly Bit (Emil Stenström) ↗ (Dec 3, 2025). AgentGid did not measure or independently verify them.

Takeaway. A measurable target like a conformance suite lets an agent iterate its way toward correctness.
AgentGid's take

This suits a solo developer or small team that is comfortable in Python and already has a strong public test suite to aim at; the case is rated advanced, and it took a full restart plus profiling and fuzzing. Watch the credit budget: Copilot's Pro plan ($10/mo) includes 1,500 monthly AI Credits, and long agent-mode iterations draw on them. Zed (Free + $10/mo, open source) lets you use your own API keys or external agents.

The agent used here

Similar use cases

Guides