links: site | repo | project page
AI Blueprint 1.13.0 is out. The main addition is simple: when an agent finishes something I can actually test, it should give me a small, concrete way to try it.
Passing tests and seeing the result myself are different things. I don’t want every handoff to end with a test count and leave me figuring out what to open, run, or look for. I also don’t want made-up manual checks for a documentation change that has nothing useful to demonstrate.
The new tryGuide: "when-testable" setting handles that distinction. It’s the
recommended default for new projects. A safe, observable change gets a short
“Try it now” section. That might be a browser interaction, but it can also be
a CLI command, API request, library call, or generated file to inspect. Invisible
internal changes don’t need an invented demonstration.
I extended the existing /check guide behavior instead of adding another
overlapping skill. The previous manual, always, and when-user-facing
settings are still supported.
The important boundary is acceptance. Giving me instructions doesn’t mean I
followed them, and the agent can’t mark the feature accepted just because it
wrote a convincing checklist. Verify and /check still cover agent verification.
The demonstration is for me.
The packed release passed a live-agent evaluation with all 26 assertions, alongside the installer and adapter checks. Version 1.13.0 is published on npm and GitHub, and the matching public docs are live.