fable5-delegate: escalate to Fable 5, fall back to Opus 4.8

# fable5-delegate (minimal). Escalate a hard task to Claude Fable 5 with a
# transparent fallback to Claude Opus 4.8. Requires: anthropic>=0.116 and
# ANTHROPIC_API_KEY. The full hardened version (automatic client-side degrade,
# streaming, cost readout, offline self-test) is downloadable at:
#   enapragma.co/resources/fable5-delegate/fable5_delegate.py
import anthropic

def delegate(prompt, effort="high", max_tokens=16000):
    client = anthropic.Anthropic()
    msg = client.beta.messages.create(
        model="claude-fable-5",
        max_tokens=max_tokens,
        messages=[{"role": "user", "content": prompt}],
        # Depth control. Do NOT send a thinking param (always-on on Fable 5;
        # disabling it 400s) and do NOT send temperature/top_p/top_k (removed on
        # Fable 5; sending them 400s).
        output_config={"effort": effort},
        # Opt into recovery so a classifier false-positive on benign work does not
        # just fail: a refused request is re-served by Opus 4.8 in the same call.
        betas=["server-side-fallback-2026-06-01"],
        extra_body={"fallbacks": [{"model": "claude-opus-4-8"}]},
    )
    # A refusal is HTTP 200 with stop_reason == "refusal" and an empty content
    # array. Check it BEFORE reading content, or you index into an empty list.
    if msg.stop_reason == "refusal":
        return None, msg.model  # the whole chain declined
    text = "".join(b.text for b in msg.content if getattr(b, "type", "") == "text")
    return text, msg.model  # msg.model tells you which model actually served it

if __name__ == "__main__":
    import sys
    answer, served_by = delegate(sys.argv[1])
    print(answer if answer is not None else "[refused by the full chain]")
    print("served by:", served_by, file=sys.stderr)