fable-check

name	fable-check
description	Decide whether a task is worth running on Fable 5 (vs Opus 4.8 / Sonnet 5) and how to run it optimally — effort level, task spec, refusal fallbacks. Use when the user asks "should this run on Fable", "is Fable worth it", or how to get the most out of Fable 5.

Fable 5 usage check

Read the plugin's data/pricing.json for current prices — resolve in order: ${CLAUDE_PLUGIN_ROOT}/data/pricing.json; if that variable is unset, ../../data/pricing.json relative to this SKILL.md; if neither exists, references/pricing.json beside this SKILL.md (a vendored snapshot — check its cached_date and flag prices as possibly stale if it is more than 60 days old). Derive the Fable-vs-Opus and Fable-vs-Sonnet cost ratios from those rates — never quote ratios from memory. Answer two questions about the task at hand: is Fable worth it here, and if yes, how should it be run.

Is Fable worth it?

Fable 5's gains are on work above what prior models can do. Route to Fable when the task is:

• Long-horizon autonomous work — overnight runs, large multi-file migrations, complex refactors expected to complete without human correction
• Problems Opus 4.8 already failed on — a concrete failure is the strongest signal
• Deep research / hardest reasoning — multi-source synthesis, ambiguous problems needing judgment
• Heavy parallel sub-agent orchestration — Fable reliably sustains long-running sub-agent coordination

Do NOT route to Fable for: routine coding, tasks with well-known solutions, anything Sonnet 5 at high effort handles (it approaches Opus 4.8 there), or security-analysis-heavy work (Fable's cyber classifiers refuse much of it — Opus 4.8 is the better tool there).

Caveats to surface every time Fable is recommended

• Refusals: safety classifiers may decline cyber/bio-adjacent requests with stop_reason: "refusal" (HTTP 200). Fallback: rerun on Opus 4.8 (API builders: use the fallbacks beta or replay history as-is).
• Long turns: single requests can run many minutes at high effort — plan timeouts/streaming/progress UX; don't block a UI on one request.
• 30-day data retention required — not available under ZDR; org retention below 30 days means every request 400s.
• Thinking is always on — omit the thinking param entirely (explicit disabled 400s).

How to run it optimally

1. Full spec up front. One well-specified first turn (goal, constraints, what "done" looks like) beats drip-fed instructions — Fable's long-horizon coherence comes from planning against a clear goal.
2. Sweep effort — don't default to xhigh. Fable at low/medium often beats older models at max. Use high/xhigh for the hardest agentic work only; low/medium for routine tasks. This is the main burn/cost lever.
3. De-prescribe migrated prompts. Step-by-step scaffolding written for older models reduces Fable output quality — state the goal and constraints, not the steps.
4. Let it delegate. Encourage parallel sub-agents for independent workstreams; give it a memory surface (even a plain .md notes file) for multi-session work.
5. Ground progress claims on long runs: require it to verify claims against tool results before reporting.

Standing recommendation for this setup

The user's working posture: Opus 4.8 as daily driver; Fable 5 escalated per-portion, then back down.

• The default escalation path is /polytropos:architect, not a session switch: Fable does the planning/meta-work once and emits an execution kit (task briefs, model-pinned subagents, guardrails, verification loops); /polytropos:execute then runs it on Opus/Sonnet at near-Fable quality. Blocked tasks escalate back to Fable one at a time. Use /model fable for a whole session only when the entire session is Fable-class work.
• Global default in ~/.claude/settings.json should be opus; if it is still pinned to a Fable model with a standing xhigh effort, offer to change it. Set effort per task, not globally.
• Once Fable 5 moves off the subscription (pay-per-token): same posture, but the architect pattern also becomes the dollar-optimal one — Fable spend concentrates in the short planning phase while execution runs at Sonnet/Opus rates.