Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
16 commits
Select commit Hold shift + click to select a range
846ac98
fix(studio): analyse migration routes on a background app-server driver
Sep 20, 2026
13d3553
fix(studio): judge migration batches on a dynamic-tool app-server turn
Sep 20, 2026
db32fe4
fix(studio): keep migration analysis, questions and delivery on app-s…
wujiaming-ai Sep 21, 2026
b0ef961
fix(studio): render migration Codex output as the intelligent build's…
wujiaming-ai Sep 22, 2026
6e32714
fix(studio): land the migration delivery turn on the first attempt
wujiaming-ai Sep 22, 2026
20427b5
fix(studio): give the migration feed the app-server's own row fields
wujiaming-ai Sep 22, 2026
efbe842
fix(studio): keep the migration question window working on 3.10
wujiaming-ai Sep 22, 2026
f03d725
fix(studio): report each migration turn's own cost
wujiaming-ai Sep 22, 2026
ce0aba3
fix(studio): run the migration turn on the Sandbox Codex app-server
wujiaming-ai Sep 22, 2026
ea5d16d
test: tell parallel tool calls from serial ones by overlap, not by st…
wujiaming-ai Sep 22, 2026
0e89974
fix(studio): hold a migration turn open until the CLI contract passes
wujiaming-ai Sep 22, 2026
46e5160
fix(studio): take the migration analysis as the model wrote it
wujiaming-ai Sep 23, 2026
ddc49c5
fix(studio): package a delivery whose CLI never got to
wujiaming-ai Sep 23, 2026
aa65e50
fix(ci): give the migration files the license header CI wants
wujiaming-ai Sep 23, 2026
8158f97
feat(studio): move the build and migration models on to 2.1 pro 260915
wujiaming-ai Sep 23, 2026
40abd6d
fix(studio): leave the BytePlus development list alone
wujiaming-ai Sep 23, 2026
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
19 changes: 16 additions & 3 deletions frontend/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -600,9 +600,22 @@ See [deployment and operation](service/studio_release_notifier/README.md).
the framework, entry point, and open questions. Structured frameworks run the
preinstalled `ak migrate`; Dify and Any projects run
`ak migrate --execution in-place` with Codex in the same Session. Evaluation
deploys a temporary Runtime, checkpoints per-case execution as JSONL, judges
batches in one fresh resumable Codex thread, and always reconciles Runtime
cleanup before completing or cancelling. Reports show 0–100 display scores,
deploys a temporary Runtime, checkpoints per-case execution as JSONL, and
judges batches in one fresh resumable Codex thread. Analysis and judging both
deliver their contract through an app-server dynamic tool instead of parsing
free text, and the analysis turn runs on a Studio background worker so a long
analysis never waits inside the upload request. Each judged batch is a durable
request the runner writes into the Session: Studio answers it with one
app-server turn carrying the verdict, and the runner validates and caches that
verdict before the next batch. A request stays on disk until it is answered, so
a Studio restart replays the same batch; a request that cannot be answered
falls back to `codex exec` inside the same batch budget. The request declares
how long the runner will listen, and the turn is sized to answer inside that
window, so even a judge batch that is too slow comes back as an answered
failure rather than a timeout. Setting
`AGENTKIT_MIGRATION_JUDGE_APP_SERVER=0` pins that scripted judge for the
whole run instead of writing judge requests at all. Cleanup is always
reconciled before completing or cancelling. Reports show 0–100 display scores,
execution success, evidence coverage, N/A counts, low-scoring and failed
cases, versions, evidence severity, and cleanup status without a pass/fail
verdict. Each raw judge score is rounded half up to a 0–100 integer before
Expand Down
Loading
Loading