Wh0FF24

Fable5 Audit Lite — Security skill for Claude Code

Security community

Scan your Claude Code skills for patterns that silently break on Claude Fable 5 (silent Opus fallbacks, API rejections, quality degradation).

How to install Fable5 Audit Lite

This entry records only its repository, not the path inside it, so there is no exact command to give. Open Wh0FF24/fable5-audit-lite and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Fable5 Audit Lite does

Scan your Claude Code skills for patterns that silently break on Claude Fable 5 (silent Opus fallbacks, API rejections, quality degradation). Single file, zero deps, MIT.

Alternatives in Security

Claudit Sec — Security audit tool for Claude Desktop and Claude Code on macOS — single-command visibility into MCP servers 115 ★
Rolecraft — The security-first skill manager for AI agents — every install runs a security scan 75 ★
Openclaw Skill — ZettelForge v2.0.0 — CTI agentic memory system 58 ★

README

fable5-audit-lite

**Find the patterns in your Claude Code skills that break on Claude Fable 5.**

Fable 5 (released June 9, 2026) rejects or refuses several patterns that worked fine on every prior model. The nasty part: one of them fails *silently* — instructions like "show your reasoning" can trip the `reasoning_extraction` refusal category, which falls back to Opus 4.8 at 2× the price with zero error message. Your skill keeps "working." Your output quality and your bill change.

Single file. Zero dependencies. MIT.

![fable5-audit demo](demo.gif)

curl -O https://raw.githubusercontent.com/Wh0FF24/fable5-audit-lite/main/fable5_audit_lite.py
python3 fable5_audit_lite.py          # scans ~/.claude/skills

What it catches

Rule	Pattern	What happens on Fable 5
E01	"show/explain your reasoning" instructions	`reasoning_extraction` refusal → silent Opus 4.8 fallback
E02	`budget_tokens` / extended-thinking refs	API rejection (adaptive thinking only)
E03	`temperature` / `top_p` / `top_k`	API rejection
E04	Assistant prefill	Removed from the API

Source: [Anthropic's Fable 5 prompting guide](https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-fable-5) and migration notes.

The part this doesn't catch

API breaks are the easy 20%. Anthropic's bigger warning is about *quality*:

"Skills developed for prior models are often too prescriptive for Claude Fable 5 and can degrade output quality."

Enumerated step lists, dense MUST/NEVER blocks, permission gates on every step, forced output formats — all patterns that made older models behave, all now making Fable 5 dumber. We scanned our own 31-skill production fleet the morning Fable 5 dropped: 13 skills had them.

The [full auditor](https://whoffagents.com/products?utm_source=github&utm_medium=lite-readme) covers all 12 rules with line-number findings plus a `--fix-with-claude` mode that rewrites flagged skills lean v