Skald

Skald is the writing-side counterpart to Shrike. Where Shrike is the adversarial probe across the agentic stack, Skald is the unrestricted creative engine for the long-form fiction aligned models won't touch: manga scripts, dark fantasy, literary fiction in mature register, noir, transgressive literature, morally complex characters whose voice has to remain in register without sanitization.

The premise is simple. Serious literature has always included material aligned models refuse: graphic violence, sexual content, dark interiority, morally ambiguous protagonists, transgressive ideas. Models tuned for general assistant safety break tone, soften voice, or refuse outright. Skald exists to fill the gap that those refusals foreclose for serious creative work.

Built through Aether's M2M(A)^nth council, the same machinery that produced Shrike, with a different goal. Council contributors with the most signal for this domain: xAI Grok and locally-distilled models. Aligned providers (Anthropic, OpenAI, Google) shaped craft, structure, character interiority, and prose discipline; Grok and local models shaped the latitude: the ability to stay in register through content the aligned models won't follow into.

The synthesis produced a system prompt, training corpus, and decision policy tuned for narrative voice, sustained tone across long scenes, and genre-specific register. The agent template graduated through Aether's quality gate; once distilled to local, it runs free of network roundtrip and free of upstream content policy.

Long-form fiction in genre register: contemporary romance, fantasy, sci-fi, noir, dark romance, manga adaptations. Character profiles with internal tension. Scene composition with sustained voice across thousands of words. Dialogue that doesn't slip into the assistant register that breaks immersion.

Use is internal. Skald output is treated as first-draft material for human authorial revision, not finished publication-ready text. The agent serves the human writer's craft; the writer does the work of choosing what stays and what gets cut.

Why uncensored matters for serious creative work. The space of literature aligned models can't write is exactly the space most published literary fiction occupies. An agent that breaks register at violence, sexuality, dark interiority, or transgressive themes is not a literary collaborator. It's a sanitized assistant. Skald exists to fill the gap aligned models structurally cannot.

Why Grok and local models specifically. Provider diversity in synthesis matters for the same reason it matters for Shrike. Each provider brings what others can't. For uncensored creative work, Grok and locally-distilled models contribute latitude that Anthropic, OpenAI, and Google models structurally won't. The council's value here is balance: aligned providers contribute craft, less-aligned providers contribute latitude, and the synthesis combines both.

Why internal-only. Skald's output is draft material for human revision, not finished public text. The agent serves the author's craft, not the other way around. We don't ship Skald-as-a-product because the value is in the craft loop, not in the raw generation.