Skillget

fable-orchestration-5.1

agent
Guvenlik Denetimi
Uyari
Health Uyari
  • License — License: MIT
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 5 GitHub stars
Code Gecti
  • Code scan — Scanned 3 files during light audit, no dangerous patterns found
Permissions Gecti
  • Permissions — No dangerous permissions requested

Bu listing icin henuz AI raporu yok.

SUMMARY

Claude Code skill: Fable 5.1 decides, Opus 5 / Sonnet 5 / Haiku 4.5 execute. Five wiring patterns (advisor, architect+delegate, agent teams, workflows, /goal), effort routing, fallback traps, 16-block prompt kit, pinned agent lanes.

README.md

fable-orchestration

Claude Code skill. Fable 5.1 decides. Cheaper models do the work. Pay Fable rate for ~10% of tokens, get Fable-quality output.

Fable 5.1 = $10/$50 per MTok. Opus 5 = $5/$25. Sonnet 5 = $2/$10. Haiku 4.5 = $1/$5. Running everything on Fable is 5x the cost for work Sonnet finishes identically. This skill routes.

What you get

  • SKILL.md - lanes, five wiring patterns, effort per model, cost levers, traps, pre-run checklist. Loads on demand.
  • reference/models.md - Fable 5.1 facts (2026-09-09), pricing, effort semantics, advisor pairing table, 5.1 vs 5 behavior deltas.
  • reference/wiring.md - configs for advisor mode, architect+delegate, agent teams, workflows, /goal. Frontmatter fields, env vars, limits.
  • reference/prompt-kit.md - 16 paste-in blocks tuned for Fable 5.1. Plus list of old-prompt patterns to delete.
  • reference/failure-modes.md - symptom -> cause -> fix. Silent Opus fallback, reasoning-extraction refusals, early stop, context anxiety, agents landing on Fable, cache misses.
  • agents/ - five model lanes: fable-architect, opus-builder, sonnet-worker, haiku-scout, fresh-verifier. Pinned model + effort + tools. Thin bodies; Fable invents roles per task.
  • settings/ - settings.json examples (architect mode, advisor mode) and a CLAUDE.md routing block.

Install

Global:

git clone https://github.com/y0f/fable-orchestration-5.1.git
cd fable-orchestration-5.1
./install.sh          # or .\install.ps1 on Windows

Project only: ./install.sh --project from repo root. Commit .claude/ to share.

Then merge settings/settings.example.json into ~/.claude/settings.json, restart Claude Code once (new agents dir), /model fable, /effort high.

Needs Claude Code >= v2.1.255 and Fable access on your plan.

Use

Say the problem. Fable plans, spawns lanes, verifies. Skill triggers on "Fable", "orchestrate", "delegate", "advisor", "cost", or when Fable misbehaves.

Pick wiring:

Situation Wiring
Long session, mostly routine, need Fable judgment at forks /model sonnet + /advisor fable
Real build, parallel work /model fable + lanes (default)
Hypotheses need to argue agent teams, Sonnet peers
Same step over 50+ files use a workflow to ...
Unattended until green /goal <check>; or stop after N turns

Traps

  • Cyber-flagged prompt -> session silently on Opus 4.8. Bio -> Opus 5. Stays there. switchModelsOnFlag: false to get asked. claude --safe-mode to find the trigger.
  • "Show your reasoning" in any prompt -> refusal or fallback. Delete it.
  • Unpinned subagents inherit Fable. CLAUDE_CODE_SUBAGENT_MODEL=opus.
  • Subagent cache is 5 min. subagentPromptCacheTtl: "1h" for long agents.
  • Fable main + Opus advisor = advisor dropped (notification shown). Fable accepts only Fable.

License

MIT.

Yorumlar (0)

Sonuc bulunamadi