Back to directory

fable-skills

Six Claude Code skills that encode behavioral discipline for the Opus 4.8 model, covering how it reports progress, when it stops work, what it touches and what it claims, each tested against real Opus 4.8 sessions with the failing transcripts included.

fable-skills

fable-skills is a set of six Claude Code skills intended to make the Opus 4.8 model behave more like its larger sibling model on the instructable parts of quality: what it claims, when it stops, what it touches, and how it reports.

Each skill targets one behavior, such as leading with the answer instead of preamble, finishing reversible in-scope work instead of asking for permission, claiming only the verification level actually reached, keeping changes to what was asked, matching a file's existing style, and spending context only on what changes the next action. Each skill was tested against real Opus 4.8 subagent sessions, with failures captured verbatim and rewritten until they no longer occurred; the transcripts are included in the repository.

Installing copies six folders into the Claude Code skills directory, plus an optional activation block for a global configuration file. It is licensed under MIT.

More in Agent skills

See all 581
  • Provenance gate

    github.com

    A deterministic authorization gateway that sits between an AI agent and its tools, tracing where each parameter of a side-effect call came from, and blocking the call or asking a human when the source does not check out.

  • A markdown framework of system rules that restates Isaac Asimov's Four Laws of Robotics as a set of prioritized loss calculations for large-language-model agents to follow, published as a discovery file for agents and crawlers to read.

  • Vitruvius

    github.com

    An engineering research agent that runs a discover, read, synthesize, verify and review loop over technical questions, reading sources directly rather than inferring from titles, and publishing its own scored verifier benchmark, weaknesses included.