Skip to content

Merge dev into main: review presets use model tier aliases - #145

Merged
factory-nizar merged 4 commits into
mainfrom
dev
Sep 28, 2026
Merged

factory-nizar merged 4 commits into
mainfrom
dev

Conversation

@factory-nizar

Copy link
Copy Markdown
Contributor

Promotes dev to main for the v9 release. Clean merge (git merge-tree reports no conflicts; every non-merge commit on main is already in dev).

Important

Do not merge until the Droid CLI release that includes Factory-AI/factory-mono#22502 is live. The new presets pass --model openai-latest-balanced / oss-latest-balanced, and today's production CLI does not know those aliases yet. On an older CLI, droid exec fails with Invalid model, and the action retries with the org's default model and posts a confusing "not a recognized model id" note on the PR. The factory-mono PR merged to dev on 2026-09-26; as of 2026-09-28 it is not on main yet. To check before merging, run droid exec --model openai-latest-balanced "hi" on the latest CLI.

Included

  • feat(review): default review presets to model tier aliases #144 feat(review): default review presets to model tier aliases (Part of AUT-2375)
    • deep now uses openai-latest-balanced with high reasoning. It resolves to GPT-5.6 Sol today (same as the current default) and moves to GPT-6 Sol once that model is generally available.
    • shallow now uses oss-latest-balanced (GLM-5.3; previously pinned to GLM-5.2).
    • The client-side policy pre-check lets tier aliases through. The CLI skips alias candidates that org policy blocks and enforces policy on the resolved model.
    • Model blocked by organization policy from the CLI now triggers the existing retry-without---model fallback.
    • README, GitLab guide, workflow template, and input descriptions steer users to review_depth. Model overrides moved to an "Advanced: model overrides" section that lists all 9 tier aliases.

After merge

Create the v9 GitHub release on main (v8 is at #143).

factory-nizar and others added 4 commits September 25, 2026 10:57
Review presets now use Factory model tier aliases (deep:
openai-latest-balanced, shallow: oss-latest-balanced) so reviews follow
the current recommended model without a droid-action release or a
workflow edit. Tier aliases skip the client-side policy pre-check (the
CLI enforces policy on the resolved model), and a CLI-side policy
rejection now triggers the existing retry-without-model fallback.

Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
Docs, the GitLab template, and action input descriptions now recommend
review_depth as the only model setting. Model override inputs move to a
collapsed "Advanced: Model Overrides" section that lists only tier
aliases, and the list of concrete model IDs is removed. Model fallback
hints point back to the default instead of the full model list.

Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
feat(review): default review presets to model tier aliases
@factory-nizar
factory-nizar marked this pull request as ready for review September 28, 2026 20:25
@factory-nizar
factory-nizar merged commit 02f8142 into main Sep 28, 2026
9 checks passed
@factory-droid

factory-droid Bot commented Sep 28, 2026 •

Copy link
Copy Markdown
Contributor

Droid finished @factory-nizar's task —— View job

Note

The requested model is not a recognized model id, so Droid retried with your organization's default model. Remove the model input (e.g. review_model) to use the recommended default, or set it to a model tier alias approved by your organization.


No inline comments were approved because the supplied PR diff is empty. The candidate describes a real standalone security-action configuration mismatch, but its line anchor is not postable.

0 inline comments posted

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant