DevEx Roadmap
roadmap.flisk.si: a community-edited map of 376 developer tools, 44 roles and 400 agent skills, drawn eight ways. Live GitHub numbers, a public checklist, and a Findings page of what the data says.
| View | What it shows | Sub-tabs |
|---|
| Compass | Easy to use (left → right) against whether it is made for small, simple setups or big, real systems (bottom → top) | One per family |
| Adoption Radar | Adopt, Trial, Assess or Hold, from release and maturity signals | One per family; a family splits into its categories |
| Lab | Every tool as an element in a periodic table | One per family |
| Journeys | Seven levels from Package Wrangler to Platform Architect, as a metro line | Backend, AI & agents, Platform |
| Roles | 44 jobs (SRE, Platform, Forward Deployed, Benchmark Engineer…) and the tools each one needs: must have, every day, good to know | One per group of jobs |
| Skills | 400 free AI agent skills from GitHub, each tested by Haiku: does it get picked for the right request, and is it followed right | One per kind of skill |
| Portfolio | Rate the tools you know 0–3: get your best-fit role, your skill shape and a page to share (a link, a download, or pages.flisk.si) | One per family |
| Ask | Type yarn vs npm vs bun and get a side-by-side answer from the data, no AI call. Hand it to your own AI for an opinion | None |
Every sub-tab has its own URL, so a link opens exactly the view you shared. Hover or focus a tool for details. Click up to 5 tools, in any view, to compare them side by side; the picks stay in the URL, so a comparison can be shared.
Honest by construction
- No dot is placed by hand. A tool's Compass position is 10 points per criterion it meets in
_data/rubric.yml. Its Radar ring follows the ring rules in the same file, with an evidence link.
- Flisk publishes the site and gets no favors. It is scored by the same checklist, has no special color or size, and sits in Assess because it first shipped in October 2026.
- Flisk appears only where it works. On Journeys, Flisk is on a level only where it does the job today or part of it. The levels it does not cover yet are tracked in flisk-sh/flisk#2647.
- The first dataset was seeded in October 2026 with AI-assisted research. It will have mistakes; that is what pull requests are for.
Testing agent skills
ruby scripts/eval-skills.rb tests every skill with Haiku through claude -p and writes _data/skill_evals.json (about 400 calls, a few dollars). The test questions live with each skill in _data/skills/*.yml: two requests where the skill should be used, one that only sounds similar, and the first step the skill asks for. A stronger model wrote them from each skill's own text; Haiku only takes the test. Run one skill with --only <id>, or a smoke test with --limit 5.
Live GitHub numbers
GITHUB_TOKEN=... ruby scripts/fetch-github.rb writes stars, last release and archive status to _data/github.json and _data/skill_repos.json. CI runs it on every build and every Monday.
Fix or add a tool
See CONTRIBUTING.md. In short: edit _data/tools/<family>.yml, run ruby scripts/check-data.rb, open a pull request with a source.
Run it locally
bundle install
bundle exec jekyll serve
Or with Docker, no Ruby needed:
docker run --rm -p 4000:4000 -v "$PWD:/site" -w /site ruby:3.3 sh -c "bundle install && bundle exec jekyll serve -H 0.0.0.0"
How it is built
Jekyll on GitHub Pages, from the flisk-sh/comparer template that also powers cmd.flisk.sh.
| Path | What it holds |
|---|
_data/families.yml | The eight families: sub-tabs, Radar slices, Lab groups |
_data/tools/<family>.yml | The tools |
_data/rubric.yml | Compass criteria, Radar rings, licenses |
_data/journeys.yml | Levels, tracks and Flisk's status per level |
_data/meta.yml | The four views and site-wide text |
_plugins/view_pages.rb | Builds one page per view and sub-tab |
assets/js/app.js | Draws the charts from assets/data.json |
License
MIT. Tool names and trademarks belong to their owners.
