Writing Staffing

Each agent loads its core skills itself. Your brief adds the rest: the lane, the extra skills to tell it to load ("load /character-sim"), the files to pass, the reader effect to create, and what to leave ambiguous or unresolved. The files each kind of work needs are in /story-memory resources/story-context.md § What Each Agent Needs.

Model Selection

Run every lane on its agent's default model. Lanes that look at different angles, focus areas or characters are parallel lanes, and the different prompts are what make them independent. Change a lane's model only when it needs a capability its default lacks.

Fan-out is the exception: the same question and files sent to two model families, for a high-stakes or disputed call where their different blind spots are the point. A family is a provider: Opus, Sonnet and Fable are one family, so two Anthropic models add no diversity. Your harness's own agent tool usually reaches only its own provider, so fan out with meridian spawn -m <model>, and check the live catalog (meridian mars models list) and its prompting guidance (meridian mars models prompting) at dispatch time.

Agent Catalogs

• resources/producers.md: @writer, @brainstormer, @outliner, @style-creator, @character-sim, @creative-researcher, and @kb-lead. Open it before you spawn an agent that writes or gathers.
• resources/reviewers.md: @critic lanes (a focus area or an edit level), @continuity-checker, @reader-sim, and how many lanes a chapter needs. Open it before staffing any critique.