Fable 5-first multi-agent orchestration for Claude Code: a token-frugal chair profile with tier routing (Sonnet 5 carries the volume, Opus 5 the hard slices), chair-sized effort per worker, a Requirements Ledger, guard hooks that enforce the discipline, and automatic teammate reaping. A clarification phase grills an…
Token-frugal multi-agent orchestration for Claude Code, built for a Claude Fable 5 chair: Fable plans and decides, Sonnet 5 carries the volume, Opus 5 takes the hard slices and escalations — each worker's effort sized by the chair — and a two-tier verification valve guards the close. When the Fable limit runs dry, an…
Runs before the agent uses a tool for ^(Agent, Task, Workflow and TaskCreate)$ tool calls, executing ledger_guard_spawn.py via python3. From Rylaa/fable5-opus5-orchestrator.
Runs after a tool call finishes for ^(Agent and Task)$ tool calls, executing agent_watchdog.py via python3 with --record. From Rylaa/fable5-opus5-orchestrator.
Runs when you submit a prompt, before the agent sees it, executing remind_reply_shape.py, agent_watchdog.py and context_watch.py via python3 with --surface (3 commands). From Rylaa/fable5-opus5-orchestrator.
Clarification protocol — grill an ambiguous request into a Requirements Ledger before any delegation. The chair MUST load this whenever a request arrives carrying ambiguity that would change the work; Rule 0.5 in the core profile only summarizes it.
Orchestrator playbook — the full delegation contract (research pipeline, subagent output contract, spawn economics, forks, teammate lifecycle, verification procedure, chair hygiene). The chair MUST load this before its first delegation of every session; the core only summarizes it.
Multi-model fusion for Claude Code + Codex: a parallel panel (1x Claude Fable 5 via claude -p locked at --effort max + 1x GPT-5.6 Sol via codex locked at ultra; /fusion-max adds a 2nd Fable 5 panelist) answers a task independently and blind, then GPT-5.6 Sol (codex, ultra) judges and a Claude Fable 5 synthesizer (via.
Claude Code instructions for Rylaa/fable-GPT-5.6-fusion, covering claude fable 5 — system prompt, claudebehavior, productinformation, refusalhandling and legalandfinancialadvice.
Answer a hard task by fanning it out to a PANEL of two frontier models running in parallel and blind — 1× Claude Fable 5 (claude -p, locked max) + 1× GPT-5.6 Sol (via codex, locked ultra), each answering the task verbatim with web search and bash, none seeing the other's work. Then GPT-5.6 Sol (via codex, ultra)…