Evaluate security upgrades through completed local commits - #98566
Draft
devjiwonchoi wants to merge 3 commits into
Draft
Evaluate security upgrades through completed local commits#98566devjiwonchoi wants to merge 3 commits into
devjiwonchoi wants to merge 3 commits into
Conversation
Contributor
Failing test suitesCommit: 9dfaa8c | About building and testing Next.js
Expand output● next-server-nft › with output:standalone › should not trace too many files in next-server.js.nft.json Other failing CI jobs |
devjiwonchoi
added this pull request to stack #98567
September 11, 2026 13:31
devjiwonchoi
force-pushed
the
jiwon/agent-upgrade/workflow
branch
from
September 11, 2026 14:16
3a464f7 to
4502a77
Compare
devjiwonchoi
force-pushed
the
jiwon/agent-upgrade/evals
branch
from
September 11, 2026 14:16
a43c8e1 to
0cbcbaa
Compare
devjiwonchoi
removed this pull request from stack #98567
September 11, 2026 14:32
devjiwonchoi
changed the base branch from
jiwon/agent-upgrade/workflow
to
jiwon/agent-upgrade/evals-base
September 11, 2026 14:33
This was referenced Sep 11, 2026
Installing a safe version does not complete a migration, and replacing Next can erase its installed guides. Turn the resolved target into retained context and a handoff to the current agent. Copy bundled workflow files and download relevant canonical guides, using Pages Router paths for versions through 13. Retain evidence outside app dependencies. The workflow requires codemods, contextual repairs, verification and commit checkpoints; experimental dry-run stops at local commits. Introduce the dry-run flag with its first execution consumer; terminal background launch follows in the next checkpoint.
Terminal users need upgrades to proceed without occupying the terminal, while an existing agent should retain its session. Detect installed harnesses, offer the CLI picker and launch detached Codex exec or native Claude background mode only when needed. Show session, logs, stop and continuation instructions, distinguishing Codex resume from live attachment. Read final startup events after observing process exit so a fast successful task still produces a receipt; preserve nonzero-exit failures and never relaunch automatically. New workers select Luna or Haiku, and existing sessions retain their model and permissions.
devjiwonchoi
force-pushed
the
jiwon/agent-upgrade/evals
branch
from
September 11, 2026 23:38
0cbcbaa to
f54bbba
Compare
devjiwonchoi
changed the base branch from
jiwon/agent-upgrade/evals-base
to
jiwon/agent-upgrade/setup
September 11, 2026 23:38
Installing the target version or printing a handoff does not establish that an agent finished the migration. The workflow needs behavioral coverage of its terminal and existing-agent entrypoints, required repairs and stop conditions. Add an isolated Vercel Sandbox suite for same-major upgrades, major migrations, existing upgrade PRs and failed PR lookup. Run all four scenarios with Codex/Luna, Codex/Terra, Claude/Haiku and Claude/Sonnet. Use real candidate tools, controlled advisory/provider responses, ordinary user prompts, withheld runtime assertions and retained transcripts and commit evidence. Observe background completion and publication attempts rather than treating launcher exit as success. Keep the public documentation benchmark unchanged and give each experiment its own result archive. Pin Next's existing lru-cache build dependency so adding the Sandbox SDK cannot change its precompiled runtime through transitive hoisting. Live provider publication, desktop attachment and eval CI automation remain outside this suite's local-commit acceptance boundary.
devjiwonchoi
force-pushed
the
jiwon/agent-upgrade/evals
branch
from
September 11, 2026 23:43
f54bbba to
9dfaa8c
Compare
devjiwonchoi
force-pushed
the
jiwon/agent-upgrade/setup
branch
from
September 12, 2026 07:19
22b79bd to
4723ad0
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Stacked on #98562.
Why?
A successful handoff or dependency installation does not prove that an agent completed an upgrade. The workflow needs behavioral coverage for migration repairs, complete local commits and prerequisites that should prevent changes.
How?
Add an independent Vercel Sandbox eval suite for same-major upgrades, major migrations, existing upgrade PRs and failed PR lookup. Run the candidate CLI and codemod with real Codex or Claude agents, ordinary user prompts, controlled advisory/provider responses and withheld behavioral assertions.
Check completed commits, async request isolation, image-quality compatibility, compatible React dependencies and local-only delivery with
--experimental-agent-dry-run. Retain transcripts and evidence separately for Luna, Terra, Haiku and Sonnet experiments. Keep the public documentation benchmark unchanged.The suite uses a manual runner; it adds no CI workflow or test-skipping rules.