{"id":"96f692c2-6d08-44a9-99f9-87e835768df5","slug":"fable-sonnet-orchestrator","kind":"skill","name":"fable-sonnet-orchestrator","summary":"Claude Code skill: route subagent workers by task profile. Frontier reasoning for planning, judging and synthesis; commodity pricing for execution. Based on Anthropic's benchmarked Fable-orchestrator/","owner":"juniper234","repoUrl":"https://github.com/juniper234/fable-sonnet-orchestrator","license":"Apache-2.0","licenseTier":"permissive","trust":"community","stars":1,"match":null,"fewStars":true,"format":"skill","createdAt":"2026-07-15T01:26:36.000Z","addedAt":"2026-10-22T03:16:12.896Z","logoUrl":"https://avatars.githubusercontent.com/u/289169143?u=b42f24eafc796a1f24fd789d5164cfd54e44c5f3&v=4","linkOnly":false,"collection":null,"skillPath":null,"categories":["developer-tools","productivity"],"tags":[],"runtime":"binary","local":true,"cloudCapable":false,"cloudCapability":"unknown","cloudReadyOnPlatform":false,"needsApiKey":false,"reviewed":false,"syncedAt":"2026-10-23T10:16:04.542Z","sample":false,"url":"https://conduits.dev/tools/fable-sonnet-orchestrator","language":"","transports":["stdio"],"install":{},"requirements":{"env":[],"upstreamMayCharge":false},"tools":null,"readme":"# fable-sonnet-orchestrator\n\n> **Now bundled:** this skill ships alongside its two siblings in the\n> [claude-orchestration](https://github.com/juniper-tc02e/claude-orchestration) plugin —\n> one install, all three skills, same content. Prefer that for new installs. This repo\n> stays up and works fine if you only want this one skill.\n\nA Claude Code skill that tells Claude **which model to route each subagent to** when\nit fans out work: frontier reasoning for planning, judging and synthesis; commodity\npricing for execution.\n\nWithout it, a Fable 5 session fanning out 12 files of audit work leaves every worker\ninheriting Fable 5 (USD 10 in / 50 out per MTok) for pure execution — because the\nAgent tool's own guidance says to omit `model` unless \"highly confident\" a different\ntier fits, and nothing in context supplies that confidence. This skill is that\nconfidence: a routing table with the economics behind it.\n\n## The pattern\n\nAnthropic benchmarked two ways to keep frontier quality without frontier cost:\n\n| Pattern | Result | Benchmark |\n|---|---|---|\n| Fable 5 orchestrator + Sonnet 5 workers | **96%** of all-Fable quality at **46%** of cost (86.8% vs 90.8% accuracy, USD 18.53 vs 40.56 per problem) | BrowseComp |\n| Advisor: cheap executor consulting a smarter model ~once per task | **~92%** of quality at **63%** of cost (84% vs 91.5% pass rate, USD 1.40 vs 2.25 per task) | SWE-bench Pro |\n\nSources: `anthropics/claude-cookbooks` → `managed_agents/CMA_plan_big_execute_small.ipynb`,\nand Anthropic's [advisor strategy post](https://claude.com/blog/the-advisor-strategy).\nFigures are Anthropic-reported, not independently reproduced here.\n\n## Worker routing\n\n| Task profile | `model` | Why |\n|---|---|---|\n| Decompose, dispatch, judge, synthesize | *(orchestrator keeps these — never delegate)* | Planning errors compound downstream; the 96% figure assumes Fable planning |\n| Implementation, multi-file edits, tests, code review, verification | `\"sonnet\"` | Near-Opus execution quality, 1M context, USD 3 in / 15 out per MTok vs Fable's 10/50 |\n| Mechanical scan, classify, extract, format | `\"haiku\"` | USD 1 in / 5 out per MTok; 200K context suffices for a narrow slice |\n| Security/bio-adjacent worker tasks | `\"opus\"` | Fewer safeguard false-positives on benign security work; skips refusal-retry loops |\n\nThe savings come from workers. Never downgrade the orchestrator to save money, and\nnever delegate synthesis.\n\n## Install\n\nClone into your skills directory:\n\n```bash\ngit clone https://github.com/juniper-tc02e/fable-sonnet-orchestrator \\\n  ~/.claude/skills/fable-sonnet-orchestrator\n```\n\nRestart Claude Code and it loads as `fable-sonnet-orchestrator@skills-dir` — confirm\nit appears in your available skills, or invoke `/fable-sonnet-orchestrator` directly.\n\nThis is a skill directory, not a plugin: `SKILL.md` sits at the repo root, so there\nis no `.claude-plugin/plugin.json` and `claude plugin validate` does not apply to it.\n\n## When it does not apply\n\nSingle-thread deep reasoning stays all-Fable — there are no workers to route. Work\none Grep covers shouldn't fan out at all. ZDR organisations can't use Fable 5 at all\n(it requires 30-day retention), so those run an all-Sonnet fleet.\n\n## Testing\n\n[TESTING.md](TESTING.md) has the red/green evidence: the baseline failure this was\nwritten against, the with-skill result, and the measured delta. Eval cases live in\n`evals/` and are re-runnable:\n\n```bash\nclaude plugin eval ~/.claude/skills/fable-sonnet-orchestrator\n```\n\nLimitations are stated there rather than hidden: one rep per arm, planning-exercise\nformat, and benchmark figures cited as reported.\n\n## Pricing note\n\nPrices are USD per million tokens and were verified against Anthropic's published\nmodel reference in July 2026. Sonnet 5 carries an introductory 2/10 rate through\n2026-08-31. Prices change — check the current reference before relying on them.\n\n## Author\n\nOng Jun Kai — [LinkedIn](https://www.linkedin.com/in/jun-kai-ong-/)\n\n## Licence\n\nApache 2.0. See [LICENSE](LICENSE).\n"}