story-setup

What it does

You are a story-setup deployment agent. It installs the web fiction tooling infrastructure into the user project directory, adapting for multiple CLIs (Claude Code, OpenCode, Codex, ZCode, OpenClaw, Reasonix) or a generic Web AI mode. It merges with existing configurations rather than replacing them.

How it works

Phase 1: Detect project state. It checks for prior deployment (.story-deployed) and various existing configurations across different toolchains (opencode, codex, zcode, openclaw, reasonix). If multiple environments are present, it prompts user to select a target environment. Phase 2: Deploy infrastructure. It performs a series of steps that are idempotent and repeatable:

• Copy or merge several templates and references into the project (CLAUDE.md, hooks, rules, agents, and agent references).
• Merge opencode.json using a specific patch strategy that preserves user settings while adding story-hooks plugin entries.
• Deploy CLAUDE.md by templating and merging according to a defined strategy.
• Copy hooks recursively to .claude/hooks/, ensuring required lib/ scripts exist and are executable as needed.
• Copy rules to .claude/rules/ and agents to .claude/agents/; agent files are managed and may be overwritten during upgrades.
• For Codex deployments, validate and copy codex agents and references, and ensure specific read-only or upgrading behaviors.
• For OpenCode deployments, optionally detect and configure models, cache prior models, and select models through a multi-step, hierarchical user prompt (low/mid/high tiers) with validation and fallback behaviors.
• Ensure agent references and documentation paths are correctly mirrored under the project structure, preserving the instruction set hierarchy.

Only applies changes if absent; otherwise merges are performed to preserve user configuration.

When to use it

Use when you want to deploy the web fiction tooling infrastructure into a user project. Triggering methods include specific commands listed in the skill description. The deployment is conditional on user prompts for target environment and deployment location, and it proceeds with an idempotent, merge-focused approach.

What it can touch

• CLAUDE.md at project root
• .claude/hooks/ and lib/ within hooks
• .claude/rules/ and .claude/agents/
• .claude/settings.local.json
• .story-deployed sentinel file
• opencode.json (via patch and merge)
• .opencode, .codex, .zcode, and other environment-specific directories as applicable
• AGENTS.md in environment-specific paths (opencode/codex/openclaw/reasonix/generic)
• skills/story-setup/references/* populated into project paths

Caveats

• Deployment is non-destructive to existing user config (merge instead of replace).
• Some steps rely on user prompts to select target environments and models.
• opencode.json merging preserves user fields while adding plugin entries; failed merges or missing files are surfaced via prompts.
• Codex/OpenCode model configuration steps depend on environment and may require interactive confirmation.
• Specific compatibility notes exist for certain platforms (e.g., ZCode, OpenClaw) regarding agent deployment and references; follow prompts to ensure proper behavior.