Re-validate this machine's model-routing configuration against the current official docs. Propose changes only — never apply an edit without explicit user confirmation.
- Fetch
https://code.claude.com/docs/llms.txt, then the current pages on model configuration, settings, subagents, advisor, and skills. - Read
~/.claude/settings.json,~/.claude/agents/*.md,~/.claude/rules/effort-escalation.md, and this skill file. - Verify against the docs just fetched (not memory):
- every settings key still exists with the same semantics:
modelabsent ⇒ plan Default applies;fallbackModelarray;advisorModel; no persistenteffortLevel - what
default,best,fable,opus,sonnet,haikucurrently resolve to for this account, and that the fallback chain still steps one tier down per entry within the documented chain-length cap - subagent frontmatter is still valid (
model: fable,tools), and whether a higher-tier alias now exists orbestbecame valid in frontmatter - advisor pairing rules still accept the configured advisor for the models in use, and whether an advisor fallback chain has become a documented feature
- no versioned model ID has crept into any of these files
- extended-context defaults: whether any
[1m]suffix became necessary or redundant
- every settings key still exists with the same semantics:
- Report a table — item, current value, docs say, action needed — then show proposed file diffs and wait for approval before editing anything.
Recommended cadence: after every claude update that changes the minor version, or monthly.