Factory-AI/droid-action v9

on GitHub

What's Changed

Review presets follow Factory's recommended models

• review_depth presets now use model tier aliases instead of pinned model IDs, so reviews move to newer models automatically without a new action release (#144)
  • Leaving review_model empty (recommended) uses Factory's recommended review model with high reasoning, currently GPT-6 Sol
  • review_model accepts tier aliases such as openai-latest-premium or anthropic-latest-balanced to pick a provider while still upgrading automatically; all 9 aliases are listed in the README under "Advanced: model overrides"
  • Specific model IDs still work in review_model / security_model, but pin the review to that model
  • The CLI enforces org model policy on the model an alias resolves to, and skips candidates the policy blocks
• README, GitLab guide, workflow template, and input descriptions now steer to review_depth and move model overrides to an advanced section (#144)

Works on older Droid CLIs

• If the installed CLI predates tier aliases, the action uses the equivalent concrete model with no note or error (#147)
  • GitHub: an Invalid model rejection of an alias reruns immediately with the stand-in, in the same attempt
  • GitLab: prepare probes the alias before the review passes and swaps in the stand-in before the org policy pre-flight; the probe runs without GitLab tokens
• Model blocked by organization policy from the CLI now triggers the existing fallback to the org's default model, and fallback notes name the model the workflow requested (#144, #147)

Full Changelog: v8...v9