ManualGen Operation Manual Generator

What Problem It Solves

Manuals for large projects are often not a question of whether documentation exists, but whether operations, sales, and customers can actually use it. Reading all code at once and then writing documentation can lead to context overload, missing modules, UI descriptions that read like API docs, and cross-module workflows scattered across separate chapters. ManualGen reframes manual generation as a layered, batch-by-batch, machine-verified process.

How It Works

The skill progresses through L0 skeleton → L1 modules → L2 regions → L3 functions → L4 operations → L5 details, where each layer builds on structured artifacts from the previous layer instead of repeatedly re-reading source code. Key mechanisms include:
- Batch persistence: each completed batch is written to _kb/Lx_xxx/*.json immediately, enabling resumable runs after interruption.
- Knowledge graph: nodes, triples, evidence chains, and Snake cross-module workflows are connected to produce end-to-end operation guides.
- CLI hard validation: scripts such as run.py verify and run.py scan_tech check artifact existence, counter consistency, and leakage of technical details like API endpoints or parameter names.
- Permission matrix: role-by-function visibility and operation boundaries are consolidated into a user-facing manual.

Scope and Caveats

It fits generating a complete user manual from scratch, focusing on named modules, backfilling changes after code updates, or mapping business workflows in large projects. It is not a good fit for one-off questions, pure coding tasks, or single-sentence summaries. Unless the user explicitly scopes modules, ManualGen defaults to full coverage; when core-module mode is used, uncovered areas must be disclosed in an appendix.