| name | fable-sonnet |
| description | Run fable-mode execution discipline on Claude Sonnet. Routes the task to the @fable-worker-sonnet agent, whose definition carries the staged loop with step-3 verification enforced hardest (Sonnet's known gap), optionally followed by a cold @fable-verifier pass. Trigger when the user explicitly asks for thorough/systematic/"deep work" handling on Sonnet ("fable on sonnet", "stage this on sonnet", "deep work mode, sonnet"). The balanced default between Haiku (cheap/fast) and Opus (peak reasoning). Do NOT use for ordinary single-pass tasks.
|
Fable Mode — Sonnet (v3, agent-routed)
v3 change: the worker is a real agent definition (agents/fable-worker-sonnet.md)
invoked by name, not a prose briefing the model can soft-ignore. Its system
prompt carries the loop and the operational rules; this skill only routes.
If a task has one obvious correct approach and fits in a single pass, skip this
loop and do it directly.
How to run it
- Confirm
fable-worker-sonnet appears in the available agent types. If not,
fall back to inline: spawn a general-purpose Sonnet agent and pass it the
rules verbatim from agents/fable-worker-sonnet.md.
- Spawn @fable-worker-sonnet via the Task tool (
subagent_type: "fable-worker-sonnet"). Brief it with: the task, the exact output path(s),
relevant context, and the pass condition its deliverable must satisfy —
name the check, don't leave it to taste.
- For independent sub-parts, spawn multiple workers concurrently and merge.
Cap concurrency at a handful. Workers do not spawn workers.
- For high-stakes deliverables, follow with @fable-verifier, briefed with
only the spec and the artifact path — not the worker's report.
- Relay results and anything marked unverified.
