feat(policy): Sonnet 5.5 reviews low and medium risk, Opus 5.5 keeps high

[soydiloreto]

📝 What changes

The organisation's default reviewer per risk level (policy/review-policy.default.yml):

Risk	Before	After
low	claude-sonnet-5, effort low	claude-sonnet-5-5, effort low
medium	claude-opus-5-5, effort medium	claude-sonnet-5-5, effort medium
high	claude-opus-5-5, effort medium	unchanged

The policy script's tests, the README's override example and CONTRIBUTING's AI-line example follow.

💡 Why

The low level named a model a generation behind; the rule is the latest version of each model. Medium-risk changes do not need the most expensive reviewer, and Sonnet 5.5 costs a fraction of Opus. High risk (providers, signatures, deletions, templates in Offload) keeps Opus 5.5, which is where the reviews found the real majors. The maintainer's call, 30 September. A repository can still override any level in its own .github/review-policy.yml.

🧪 How I tested it

• python3 scripts/policy.py --test

🤖 AI-generated · Claude Opus 5.5 (Anthropic)

[dilux-bot]

Claude review · risk high · complexity low · type feat

This PR changes the organisation's default reviewer: claude-sonnet-5-5 now reviews both the low and medium floors, and claude-opus-5-5 keeps high. The policy self-test, the workflow header comment, CONTRIBUTING and the README override example are all updated to match, and no stale copy of the old wording is left in the repository. The README example sets effort: high, which policy.py accepts. The open thread on policy/review-policy.default.yml:31 is still unresolved. It has no replies and no evidence has been added. The newest Sonnet ID I know of is claude-sonnet-5. If claude-sonnet-5-5 is not a valid model ID, every low- and medium-floor review and all issue triage across DiluxOne would fail. Linking one successful run with --model claude-sonnet-5-5 would settle it. One minor: the new self-test case "a workflow gets the strong reviewer" repeats the inputs of the existing "a workflow is high" case. The model check could go into that case instead. The description does not mention the workflow header comment edit, but that is a comment-only change and does not change behaviour.

• major policy/review-policy.default.yml:31: Unverified model ID claude-sonnet-5-5 in the org default (open thread): if it is wrong, every low/medium review and issue triage fails at --model; link a successful run to settle it
• minor scripts/policy.py:220: New case 'a workflow gets the strong reviewer' repeats the inputs of 'a workflow is high'; move the model check into the existing case

_{Policy floor: high (touches high-risk paths: .github/workflows/claude-review.yml, CONTRIBUTING.md, README.md, policy/review-policy.default.yml, scripts/policy.py). Reviewed a9c36c2 (whole pull request; review 2 of 5 automatic). Author trusted for auto-merge: true.}

_{🤖 AI review · claude-opus-5-5 (Anthropic) · $0.14, 5 turns}