You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Reviews stay consistent from develop to QA to staging. PRISM keeps a
short-lived record (a temp file on this machine; nothing is ever written to
your repositories) of what it concluded about each change, matched by the code
that changed so it survives rebases and squashes, and gives the reviewer the
earlier findings when the same change comes back in a later PR (or a Retry). A
defect the earlier stages missed is flagged as a late find in the job log
and the PR description, since a fix made that late hasn't been through the
testing in between. Only a late Critical/High finding blocks the merge; a
late Medium or lower is reported and approves with comments. Nothing changes
if there is no earlier review of those changes.
Reviewers look harder the first time. Every review now checks failure
handling and edge cases, the impact on callers, and searches the repository
for the same defect elsewhere: each finding lists the other places the pattern
occurs (or says none were found), so one fix can cover them all. Findings can
be longer as a result, and reviews use somewhat more tokens.
Codegen tool integration (optional). In Help, tick Accept reviews from a
code-generation tool and PRISM will review the pull requests a codegen tool
opens, send "request changes" findings back to it, and merge once the PR is
approved — repeating up to five rounds before it hands over to you. Off by
default and local to your machine; a job from the tool is an ordinary job
(marked via codegen) with every existing check, including the high-impact
confirmation, still in force. Nothing changes if you never turn it on.
The manual is updated for all of the above: how to read the new finding
markers, the new safety notes and troubleshooting entries.