chore(policy): review every risk level with Sonnet 5.5, at low, medium and high effort

[soydiloreto]

📝 What changes

Every review now uses Sonnet 5.5; the risk level decides only the effort:

Risk	Before	After
low	Sonnet 5.5, effort low	unchanged
medium	Sonnet 5.5, effort medium	unchanged
high	Opus 5.5, effort medium	Sonnet 5.5, effort high

The review's fallback when the policy names no model is Sonnet 5.5. Replies to @dilux-bot and the weekly learnings, which use the high reviewer, follow. CONTRIBUTING, the README, the workflow comments and the policy self-tests stop describing a light and a strong model.

💡 Why

The maintainer's decision: one model for every review, with the effort rising with the risk.

🧪 How I tested it

python3 scripts/policy.py --test passes, including the case that a workflow change gets the high level (Sonnet 5.5, effort high).

🤖 AI-generated · Claude Opus 5.5 (Anthropic)

[dilux-bot]

Claude review · risk high · complexity low · type chore

This PR changes the reviewer for high-risk PRs from Opus 5.5 at medium effort to Sonnet 5.5 at high effort, in policy/review-policy.default.yml. The fallback in claude-review.yml for a policy that names no model is now claude-sonnet-5-5, so it agrees with the policy. Everything that wraps the policy changes with it: the self-test now checks both the model and the effort for the high level, and CONTRIBUTING, the README (callers table and example policy) and the workflow header comments no longer mention a light or strong model. I searched the repository for leftover "opus", "strong model" and "light model" wording and found none (the Opus mentions left are in AI attribution lines, which is correct). review-reply.yml, review-learnings.yml and the issue workflows read the model and effort from the policy, so they follow without edits, as the description says. high is an effort value policy.py already accepts. Risk is high because the change sets the reviewer for every DiluxOne repository; it is a small, consistent config change and nothing blocks it. The type is chore: the reviewer's settings change, nothing is added or corrected, and no caller has to change anything. The description matches the diff. I did not run policy.py --test myself: running it needed a shell command that required approval in this environment.

No findings.

_{Policy floor: high (touches high-risk paths: .github/workflows/claude-review.yml, .github/workflows/issue-triage.yml, CONTRIBUTING.md, README.md, policy/review-policy.default.yml …). Reviewed 35e2475 (whole pull request; review 1 of 5 automatic). Author trusted for auto-merge: true.}

_{🤖 AI review · claude-opus-5-5 (Anthropic) · $0.27, 10 turns}