links: site | repo | project page
AI Blueprint 1.5.4 is one of those releases that looks small from the outside, but makes the workflow feel a lot less fragile when you are actually using it.
The biggest fix was status drift. /complete marks a build-plan item as done,
which means the plan file changes after implementation. The old overview check
treated that as suspicious drift because it compared the plan too literally.
Now it normalizes line-start completion markers before hashing, so real plan
changes still matter but checking off finished work does not break the gate.
I also cleaned up the way the dashboard reads current work and history. It now accepts the documented feature heading and older numbered feature fields, so valid work does not show up as malformed just because it came from a previous format.
what shipped:
- final read-only walkthroughs for completed implementations
- clearer Efficient and Guided behavior around review cadence
- stack-neutral build and verify handling
- better
/status,/complete, and dashboard parsing - regression tests around overview drift and completion state
- updated command docs and public site pages
The decision I liked most was keeping /try as the manual product-review path.
It is tempting to turn every command into a giant catch-all, but /implement
should explain the code and /try should tell a human how to judge the result
in the real app.
That is the shape I want Blueprint to keep: more automation, but each command still has a job. The workflow can get faster and smoother without making the merge gate casual.