Skip to content

Releases: dotCMS/openrouter-code-review-action

dotbot code review v1.3.6

Choose a tag to compare

@github-actions github-actions released this 29 Sep 17:22
0172411

Auto-generated release from merge to main.

dotbot code review v1.3.5

Choose a tag to compare

@github-actions github-actions released this 29 Sep 16:58
2bbe392

Auto-generated release from merge to main.

dotbot code review v1.3.4

Choose a tag to compare

@github-actions github-actions released this 29 Sep 16:47
de93a1b

Auto-generated release from merge to main.

dotbot code review v1.3.3

Choose a tag to compare

@github-actions github-actions released this 29 Sep 16:40
e64d19f

Auto-generated release from merge to main.

dotbot code review v1.3.2

Choose a tag to compare

@github-actions github-actions released this 29 Sep 16:27
0d3d3f8

Auto-generated release from merge to main.

dotbot code review v1.3.1

Choose a tag to compare

@github-actions github-actions released this 29 Sep 16:22
3f4cfa1

Auto-generated release from merge to main.

dotbot code review v1.3.0

Choose a tag to compare

@github-actions github-actions released this 29 Sep 16:16
b1c5b6a

Auto-generated release from merge to main.

dotbot code review v1.2.0

Choose a tag to compare

@github-actions github-actions released this 04 Sep 22:35
bbe2345

Auto-generated release from merge to main.

dotbot code review v1.1.0

Choose a tag to compare

@github-actions github-actions released this 04 Sep 22:00
aa372f5

Auto-generated release from merge to main.

v1.0.3

Choose a tag to compare

@wezell wezell released this 27 Aug 03:26
a716ed1

What's New

Automatic PR approval (machine user)

  • New github_approval_token input / DOTBOT_GITHUB_USER_PAT env: when every reviewer model reports Overall: patch is correct, dotbot submits an APPROVE review as the PAT's user (e.g. dotCMS-Machine-User)
  • Idempotent per head SHA — re-runs don't spam duplicate approvals; any model dissent withholds approval; failures are non-fatal warnings

Reviewer observability + wall-clock budget

  • [openrouter-request] per-request elapsed + finish reason (live in CI logs)
  • [openrouter-tool] tool calls with readable arg summaries at debug 1
  • [openrouter-usage] cumulative prompt/completion/total tokens per model pass
  • [openrouter-reasoning] per-request reasoning char counts
  • Reviewer <model> finished in Ns per pass
  • New model_timeout_seconds input (default 900s, 0 disables): aborts a wedged reviewer pass instead of burning credits indefinitely

Full Changelog: v1.0.2...v1.0.3