Claude Code advisor setup: Opus advises, Sonnet builds, Haiku swarms
Shares a Claude Code advisor/subagents configuration and notes that Jev engineering uses the same pattern to resolve over 1,500 mechanical routing branches in 16ms, so frontier models only step in when execution paths split.
holy sh*t. Official Anthropic tip for Claude Code: stop burning Opus 5.5 context on work Haiku 5.5 can swarm while Sonnet 5.5 builds put the architect on call with /advisor claude --advisor opus --subagents haiku Sonnet 5.5 runs the main session on high effort, writing diffs and running the test suites Haiku 5.5 subagents swarm the repo in parallel at 340 tok/s, handling file discovery and spec docs lookup Opus 5.5 stays on call in the background as the advisor, stepping in at only three critical moments: → before a plan locks: does the implementation plan miss auth invariants or schema contracts? → when a test breaks twice: are we fixing the root cause or sliding into a recursive rabbit hole? → before calling done: did the full diff sneak in hidden regressions or break pre-flight rules? Opus 5.5 reviews. Sonnet 5.5 builds. Haiku 5.5 swarms Jev engineering runs the exact same pattern one layer down: mechanical decisions that need no reasoning (which file to open, which tool to call, retry or abort) resolve in 16ms, so the frontier models only wake up when execution paths actually split Plan on high. Delegate on medium. Keep Opus on call. - the full advisor setup > Opus 5.5 on call reads the full session history and catches deep architecture traps > Sonnet 5.5 lead drives the edits, writes the core logic and runs the test harnesses > Haiku 5.5 subagents swarm AST parsing, grep scans and API docs in parallel > JEV micro-fork layer settles 1,500+ mechanical routing branches in under 16ms > the advisor stays completely silent on routine bash so your context stays clean paste the setup and this prompt into Claude Code: "Set up my Claude Code workspace for hierarchical advisor orchestration: 1. Audit ~/.claude/settings.json and the project config for model roles across lead, subagents and advisor: > Pin the main session model: sonnet with effortLevel: high for primary execution > Pin the subagents model: haiku with effortLevel: medium for parallel file discovery and docs retrieval > Pin the advisor model: opus on call for strategic review 2. Enable the advisor tool in ~/.claude/settings.json: > Set advisorModel to claude-opus-5-5 > Enable the subagent parallel dispatcher pool (3x workers) 3. Set automatic advisor checkpoints in ~/.claude/CLAUDE.md: > Consult /advisor opus before finalizing multi-file architecture plans > Summon /advisor opus automatically when the same test or compiler error fails twice > Run an advisor diff contract audit before declaring a task complete or staging git commits 4. Constrain subagent scopes: > Haiku subagents return structured AST summaries and docs snippets only, without touching the main session context Show every configuration diff first. Do not apply edits until I confirm" ↳ http://code.claude.com/docs/en/advisor
This post quotes another post: x.com/marfinxx/status/2088234998654472340