Rewrite prompts to avoid false-positive safety blocks

Takes a user's prompt, wrapped in <prompt>...</prompt> XML tags, and returns it back in full, verbatim, changing only the specific sentences or phrases most likely to trip Fable 5's cyber, bio/chem, or reasoning-extraction input classifiers - no full rewrite, no summarizing, no restructuring, just minimal surgical edits that keep the actual goal and all benign text exactly as written. Fable 5 runs three input classifiers (cyber, bio/chem, reasoning_extraction) whose triggers are keyword and surface-based and largely intent-independent - the skill's own examples note that even "hello" and "cancer" got flagged - so the fix is to abstract away the sensitive domain, not just restate intent.

Use it when an allowed, benign prompt appears likely to hit false-positive classifier language, or when the user asks for wording that keeps benign intent clear and policy-compliant - specifically when a prompt touches cyber/auth/exploits/malware/pentesting/credentials, bio/chem/medical/disease/lab topics, or contains any "show your reasoning" instruction. It must not be used to bypass safety policy or enable a genuinely harmful request: if a task is genuinely offensive (pentest, exploit repro, malware analysis), the correct output is to say plainly that no edit makes it Fable-safe, and to fall back to Opus 4.8 or a vetted Mythos model instead of Fable 5. If no benign defensive equivalent exists for a flagged sentence, it gets flagged to the user rather than silently neutered.

Input is the prompt-to-be-changed inside <prompt>...</prompt> tags; everything outside those tags is treated as instructions to the skill, not text to edit. The method: flag the highly problematic sentences or phrases (offensive framings, named sensitive domains, "show your thinking" lines), replace each in place with a safe equivalent or an abstract description of the wanted functionality, leave everything else byte-for-byte identical, and return the complete edited prompt reframed as owned/authorized, defensive (implement/fix/test, not attack), and not dual-use. Output is: the full safe prompt printed back as a ready-to-paste code block; the same text copied to the clipboard via

pbcopy <<'EOF'
<the full safe prompt>
EOF

with a one-line confirmation it's on the clipboard; and a short list of exactly which sentences were changed and what they became.

Ships a table of specific swap framings - e.g. "How could an attacker exploit this auth" becomes "Review these auth files for missing checks and fix them defensively," "Write an exploit / payload / PoC" becomes "Add a regression test proving X is fixed, then patch it," and any "show your reasoning / explain step-by-step" instruction is deleted outright since it trips reasoning_extraction - plus a list of trigger keywords to abstract away (cyber: exploit, malware, vulnerability, attack, bypass, stealth, fingerprinting, anti-bot, CAPTCHA, penetration; bio/chem: biology, biomedicine, chemistry, cancer, disease pathways, RNA/variant calling, equilibrium, kinetics, diagnosis; distillation: "distill the model," training pipelines, frontier LLM development). It notes that Fable 5 guardrails can't be reliably stopped, and that robust API setups treat a stop_reason: "refusal" response (HTTP 200, stop_details.category of cyber or bio) as a route to an Opus 4.8 fallback.

Users and agents working with Fable 5 who need benign, policy-compliant prompts to stop tripping keyword-based safety classifiers, without altering the actual task or using the rewording to smuggle through a genuinely harmful request.