The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the agent to install the novel-ralph-skill package using the uv tool install command. This package contains the necessary CLI tools for managing the novel's state and manuscript structure. These are vendor-provided resources intended for use with this skill.
[REMOTE_CODE_EXECUTION]: The setup instructions involve installing and executing code from the novel-ralph-skill package. This process installs the novel binary to the user's environment, which is subsequently used to manage all phases of the writing process.
[COMMAND_EXECUTION]: The skill relies on frequent execution of the novel CLI tool for state manipulation, drafting progress, and compiling manuscript files. For example, it uses novel state set-chapters to pass structured chapter plans to the tool as shell arguments.
[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection because it processes and analyzes untrusted data, specifically the user's story premise and the drafted novel content. Malicious instructions embedded in these texts could potentially influence the agent during its iterative review cycles (e.g., the "Spiteful Critic" or "Knitting Circle" passes). Sanitization is handled through the structured phase-based workflow.
[DATA_EXPOSURE]: The skill manages character profiles, world-building data, and manuscript drafts within a local working/ directory. It reads from and writes to state.toml and log.md to maintain persistence across agent turns. Access is restricted to the skill's operating environment.

novel-ralph