story-review

Invoke the review with a chapter file or range, and it coordinates specialized reviewer perspectives covering story architecture, character voice, prose quality, and factual consistency. Findings are returned in a unified schema with severity levels S1 through S4 and concrete fix suggestions.

Full mode spawns four reviewer agents in parallel, lean mode spawns only the story-architect and consistency-checker, and solo mode runs the review in the current session without subagents. If required agents are missing, malformed, or fail to spawn, the review automatically degrades to solo and reports the fallback reason.

Yes, it loads dedicated rubrics for Fanqie, Qidian, and Zhihu based on the project's stated target platform, falling back to a generic web-fiction rubric when no platform is identified. If rubric files are unreadable, embedded fallback criteria are used and reported as the rubric source.

Yes, bundled Node.js scripts deterministically flag high-risk AI sentence patterns, em-dash misuse, fragmented sentences, overlong paragraphs, and model degeneration such as verbatim repetition or truncation. These script findings merge into the report as prose or format issues without modifying the source files.

Multi-chapter reviews are split into batches, each producing independent findings that are later merged. Open foreshadowing items due for resolution are injected into each batch so dangling hooks from earlier chapters are caught, and newly discovered open items are written back to the tracking file.

Fallback occurs when required reviewer agent files are missing or malformed, the deployment is stale, the agent tool is unavailable, a spawn fails at runtime, or the review is already running inside a subagent. The report header lists the exact fallback reason and suggests rerunning the setup command.