Repository navigation
chore: update model references for Claude Opus 5.5 - #351
Conversation
- Global CLAUDE.md: describe the default as the `opus` alias rather than pinning "Claude Opus 5"; rename "Working with Opus 5" to "Working Style" and drop the Opus 5-specific claim that lowering effort doesn't shorten output (the Opus 5.5 guidance is the reverse: lower effort before prompting for brevity). Widen the verification rule to cover verify-with-a-subagent phases in commands and agents. - CI review workflow: the fallback prompt said "REQUEST_CHANGES for any issues found", contradicting the central prompt's convergence policy; align it. Drop pinned model versions from workflow comments. - Model IDs: hermes default claude-opus-4-8 -> claude-opus-5-5; summarize and mcp-builder eval defaults claude-sonnet-5 -> claude-sonnet-5-5; statusline example comment. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01LBftoc1JCMLow9LDFZMADs
claude-code-action skips review on PRs that modify its workflow file, and main requires an approving review. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01LBftoc1JCMLow9LDFZMADs
![claude[bot]](https://avatars.githubusercontent.com/in/1236702?s=60&v=4){kind=small}
| LINK_RESET=$'\e]8;;\e\\' | ||
|
|
||
| # Model name — prefer the friendly display_name (e.g. "Opus 4.8") over the raw id | ||
| # Model name — prefer the friendly display_name (e.g. "Opus 5.5") over the raw id |
There was a problem hiding this comment.
[Suggestion] This file still has an old model ID: the usage example in the header comment on line 8 uses claude-opus-4-6 as the model id. Since you are updating this file's examples anyway, consider changing it to claude-opus-5-5 so it matches. It is a comment only and does not change behavior.
There was a problem hiding this comment.
Code Review — Round 1
Summary
This PR changes docs and model IDs only. The new IDs (claude-opus-5-5, claude-sonnet-5-5) are valid. Pointing the global CLAUDE.md at the opus alias instead of a named version means it will not go stale at the next release. Dropping the "lowering effort does not shorten output" line fits the 5.5 guidance. The changes in evaluation.py only touch string defaults and help text, so they do not change behavior beyond which model is called.
The workflow files still say Claude Opus 5 / Sonnet 5 (claude-code-review.yml:68, claude.yml:49). That is expected, because the PR body moves those edits to PR 352.
One small gap: the PR body table does not list the evaluation.md:599 change, where the "more capable model" example moves from Sonnet to claude-opus-5-5. The change is reasonable, since Opus is the more capable tier.
Verdict
APPROVE - No blocking findings. One non-blocking suggestion is inline.
Automated review by Claude Code
1 participant
This is the first of a series of PRs that update the Claude Code setup for the Opus 5.5 generation. It changes model references and the global CLAUDE.md only. The two workflow-file edits moved to #352, because the bot reviewer skips PRs that touch its own workflow.
Global
CLAUDE.md(loaded every session, and by Antigravity throughGEMINI.md):opusalias instead of pinning "Claude Opus 5", so the text doesn't go stale at the next release.Model IDs:
home/.hermes/config.yamlclaude-opus-4-8claude-opus-5-5summarizeskill (recommended model)claude-sonnet-5claude-sonnet-5-5mcp-buildereval script (default)claude-sonnet-5claude-sonnet-5-5statusline-command.sh(example in a comment)"Opus 4.8""Opus 5.5"Left on purpose:
docs/case-study.mdquotes Opus 4.5 pricing from Nov 2025 as a dated historical record.make checkpasses, andevaluation.pycompiles.🤖 Generated with Claude Code
https://claude.ai/code/session_01LBftoc1JCMLow9LDFZMADs