Roadmap as of 14 October 2026
What comes next for html-explainer
Every page is now checked automatically, so next we test pages with outside readers, and new page types wait until an outside user needs one.
A plan, not a promise. Items move as we learn.
- Focus: checked decision pages for readers who cannot hold a long document.
- The outside-reader test decides what comes next.
- Free and MIT licensed, and it stays that way. Ask by opening an issue.
- Now, next and laterTwelve live items in three lanes, and what waits on what.
- What unblocks the mostThe outside-reader test and the automatic check.
- What is left offThree things parked, each with the way back.
- The December reviewWhat happens on 12 December 2026, either way.
Chapter 1 of 4, about 2 minutes
What is under way, next and later?
Before this: the focus is checked decision pages.
Now
Next 4 weeks. Committed. Top item first.
- Trigger reliablySharper skill descriptions and trigger tests.
- Test with outside readersFive to ten readers read one text three ways.Fixed date: by 12 December, the review needs it.
- Settle the name, then listPick the final name, then submit to registries.After: Trigger reliably
Next
1 to 3 months. Likely. Not in order.
- Optional browser checkReports sideways scroll, split words and clipped commands.
- Tested installs in other agentsCodex, Gemini CLI, Cursor and Copilot.
- claude.ai uploadA zip per skill on each GitHub release.After: Settle the name, then list
- Publish where Claude publishesWhen it can, the agent offers an artifact link.Exploring: depends on the platform
- A one-minute demoRecorded from the gallery.
Later
3 months or more. Exploring. Not in order.
- Pull request pages in CIA GitHub Action writes one per pull request.After: Test with outside readers
- Brand themesYour own colour and type, rules still checked.After: Test with outside readers
- More languagesAfter Japanese, on request.
- Commercial supportFor teams that adopt it, if a team asks.After: Test with outside readers
Take from this: five items wait on another, and three wait on the outside-reader test.
Shipped in 1.6.0 to 1.8.0: 3 itemsChangelog
- Every page type redesigned around its own picture (1.8.0)
- The field-guide page type (1.7.0)
- Roadmap lanes that say what they mean, and prioritising that adds up (1.6.0)
Before that, 1.5.0 checked every page automatically, upgraded old pages and added answers by tapping. 1.4.0 moved each skill into a self-contained folder, and added this roadmap, a public site and an automatic check on every change.
Parked on purpose: 3 items, each with what would bring it back
Show the detail and the reasons behind the items
Upgrade old pages. A script swaps an older page's styles and reader script for the current template's. Each page carries its own copy of the design, so a page made under 1.2 keeps 1.2's phone bugs.
Trigger reliably. The tests are built with claude plugin eval and cover the phrases people actually type. In the author's own sessions about two requests in three did not visibly load the skill.
Settle the name, then list. The listing goes to Anthropic's plugin directory and the agent-skills registries. Renaming a listed plugin orphans every install, so the name comes first.
Optional browser check. check.mjs --render opens the page in a local Chrome at phone and desktop widths. It runs only when Playwright is installed. The 1.3.0 layout defects were invisible to the static checker.
Tested installs in other agents. Each agent gets a recorded install and one finished page. A Codex plugin manifest and a Gemini extension are added if those routes keep the folders intact.
Pull request pages in CI. The GitHub Action writes a pr page for each pull request and links it in a comment.
Brand themes. You swap the colour and type tokens for your own, and the contrast and meaning rules are still checked.
More languages. They come after Japanese, on request, using the same translate and merge workflow.
A one-minute demo. It is recorded from the gallery and shows skim mode, one chapter at a time, read aloud and the ballot round trip.
claude.ai upload. Skill descriptions are short enough for the upload form.
Publish where Claude publishes. When the agent can publish a page as a Claude artifact, it offers that link next to the file.
The reasons for the outside-reader test and the automatic check are in chapter 2.
Remember: twelve live items sit in three lanes, and five of them wait on another.
Chapter 2 of 4, about 2 minutes
What unblocks the most?
Before this: twelve live items sit in three lanes, and five of them wait on another.
The outside-reader test unblocks the most. Every page so far was judged by its author, and this test decides what comes next.
Five to ten readers, some with ADHD, read the same content three ways: a plugin page, a default Claude page and plain Markdown. We measure time to a correct decision and what they remember a day later.
The automatic check shipped first, in 1.5.0. Before it, the skill told the agent to run the checker, and an instruction can be skipped.
- The skill tells the agent to run the checker.
- An instruction can be skipped.
- A plugin hook runs the checker at the end of every turn that wrote a page.
- Any failure goes back to the agent before you see the page.
Take from this: a hook cannot be skipped, so every page is checked before you see it.
Remember: the outside-reader test decides what comes next, and a hook makes the check impossible to skip.
Chapter 3 of 4, about 2 minutes
What is left off on purpose?
Before this: the outside-reader test decides what comes next, and a hook makes the check impossible to skip.
Work that makes pages more trustworthy, easier to install or easier to read comes before new page types. Three things are parked on purpose.
| Item | Why it is parked | What would restore it |
|---|---|---|
| New page types | Seventeen types plus gather and atlas cover every request seen so far. More types would spread the checker thin before outsiders have tried the current ones. | An outside user brings content that no current type can hold. |
| Motion companion | Animated walkthroughs. Parked on 9 October 2026. It needs a Claude Code hook or API to drive it. | A supported hook or API that can render motion from a page. |
| A hosted service or paid tier | Pages work offline as single files, and everything that makes them useful is in public files. | Teams asking for something that cannot be a file. |
Take from this: each parked item has a reason and a stated way back.
Remember: three things are parked, and each names what would bring it back.
Chapter 4 of 4, about 1 minute
When is this reviewed?
Before this: three things are parked, and each names what would bring it back.
On 12 December 2026 this roadmap is reviewed against outside feedback and the reader test. The question is whether plugin pages help outside readers more than a default page does.
- The Later lane stays on the roadmap for 2027.
- The roadmap shrinks to maintenance.
- The plugin stays maintained for the people who use it.
Take from this: the review ends one of two ways. Later stays, or the roadmap shrinks to maintenance.
Open an issue to ask for something. A request from someone who has used a page moves an item up faster than anything else.
Remember: on 12 December 2026 the review either keeps Later or shrinks the roadmap to maintenance.
Recap
- Twelve live items sit in three lanes, and five of them wait on another.
- The outside-reader test decides what comes next, and a hook makes the check impossible to skip.
- Three things are parked, and each names what would bring it back.
- On 12 December 2026 the review either keeps Later or shrinks the roadmap to maintenance.