Skip to content

feat: upgrade PR review to Opus 4.8 and issue answers to Sonnet 4.6 (engine v1.6)#757

Merged
sudsali merged 3 commits into
masterfrom
shadow-v1.4-opus48
Jul 2, 2026
Merged

feat: upgrade PR review to Opus 4.8 and issue answers to Sonnet 4.6 (engine v1.6)#757
sudsali merged 3 commits into
masterfrom
shadow-v1.4-opus48

Conversation

@sudsali

@sudsali sudsali commented Jul 2, 2026

Copy link
Copy Markdown
Contributor

Summary

Upgrades the models the bot uses, via one engine version bump:

  • PR review → Claude Opus 4.8 (investigator + critic stages) for stronger analysis
  • Issue / follow-up answers → Claude Sonnet 4.6 for better user-facing responses
  • PR reporter stays Claude Haiku 4.5 — it only formats results into JSON, so a larger model isn't needed

Why

  • The investigator/critic stages do the actual code analysis; Opus 4.8 is a meaningful capability improvement there. The engine bump also carries the fix required to run 4.8 on Bedrock.
  • Issue and follow-up answers are user-facing prose. Internal evaluation on a set of past deequ questions found Sonnet 4.6 noticeably more accurate than Haiku for these, so it's worth the higher cost on that path only.
  • The reporter stage is JSON formatting; Haiku is well-suited and cheaper, so it's unchanged.

Changes

  • .github/workflows/issue-bot.yml — bump the shared engine pin (uses: + shadow_ref)
  • .shadow.ymlmodels.issue: us.anthropic.claude-sonnet-4-6

Rollout

  • Issue-answer model (Sonnet 4.6) takes effect on merge — it's in .shadow.yml.
  • PR-review model (Opus 4.8) is applied by setting the BEDROCK_MODEL_ID repository secret to us.anthropic.claude-opus-4-8 after merge.

Testing

Validated against historical deequ PRs and issues in dry-run mode before this change.

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.

BEDROCK_MODEL_ID is now us.anthropic.claude-opus-4-8; v1.4 carries the
sampling-param fix that 4.8 requires (4.8 hard-fails on earlier pins). Reporter
stays Haiku 4.5.

@github-actions github-actions Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No issues found.


Generated by AI (model: us.anthropic.claude-opus-4-7, prompt: 4d2f5d73) — may not be fully accurate. Reply if this doesn't help.
Reviewed by Shadow · github.com/sudsali/shadow

@github-actions github-actions Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No issues found and CI is passing. Auto-approved.


Generated by AI — human merge required.

Consolidates the model upgrades into one change:
- Engine pin v1.4 -> v1.6 (v1.6 includes the Opus 4.8 sampling-param fix and
  the issue-model wiring).
- .shadow.yml models.issue = Sonnet 4.6 for customer issue/followup answers,
  selected by an n=20 3-judge bench (14-5 over Haiku, half the hallucinations).
  PR reporter stays Haiku (JSON formatting only).
- Opus 4.8 for PR review remains applied post-merge via the BEDROCK_MODEL_ID
  secret (env > yaml).
@sudsali sudsali changed the title feat: run PR review on Opus 4.8 (engine v1.4) feat: upgrade PR review to Opus 4.8 and issue answers to Sonnet 4.6 (engine v1.6) Jul 2, 2026

@github-actions github-actions Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No issues found.


Generated by AI (model: us.anthropic.claude-opus-4-7, prompt: 4d2f5d73) — may not be fully accurate. Reply if this doesn't help.
Reviewed by Shadow · github.com/sudsali/shadow

@github-actions github-actions Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No issues found.


Generated by AI (model: us.anthropic.claude-opus-4-7, prompt: 4d2f5d73) — may not be fully accurate. Reply if this doesn't help.
Reviewed by Shadow · github.com/sudsali/shadow

@sudsali sudsali merged commit 18094c1 into master Jul 2, 2026
6 checks passed
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