A source build is only the first adoption check
rhwp combines documented HWP/HWPX viewing and editing with CLI conversion and MCP access. Our recorded source build succeeded, but we did not measure whether it preserves your documents' layout. That distinction matters when choosing a tool for forms, tables or print output.
The English README describes a Rust and WebAssembly engine with HWP 5.0 and HWPX parsing, browser rendering, editing and export. HML support is narrower: the project limits it to corpus-verified HWPML 2.9/2.91 structures and describes preservation checks that block lossy saves involving unsupported content. These are project claims, not results from our sandbox.
Choose the interface that matches the job
The README documents @rhwp/editor for embedding the editor UI and @rhwp/core for direct WASM parser and renderer calls. It also describes native CLI release archives and an embedded MCP server for hosts such as Claude Code. MCP operations include document information, search, field filling and conversion; session tools support repeated queries and edits without reparsing.
Those interfaces give developers several integration options. Our run did not exercise the browser editor, npm packages, MCP sessions or document conversion, and it did not assess how an MCP host handles document data.
The v0.8.4 release is relevant distribution history. It retained native CLI releases, the two npm packages, GitHub Pages, editor marketplaces and Chrome, Edge and Firefox extensions. It withdrew channels unintentionally enabled in v0.8.3, including PyPI, @rhwp/node, the GHCR CLI image, operating-system packages and installation automation. Treat that as a dated release policy, rather than evidence that every historical installation route remains supported.
What our recorded run established
Our run was recorded on October 9, 2026 at commit 1a76570e833917d15817415a53c09ad61ab3203f, whose commit date is October 6, 2026. The build log identifies rhwp v0.8.7. This is a different scope from the historical v0.8.4 release discussion.
The dependency-download step completed successfully in 23 seconds. The record lists 241 dependencies; this is not an installed-binary check. The build step completed successfully in 195 seconds, including the compilation activity visible in its tail.
The test command exited 101 after 209 seconds and did not time out. The decisive error was a missing generated source file, tests/generated/regression_suite_028.rs, which prevented compilation of the regression_suite_028 test target. The available record does not explain why the file was missing or provide a passing-test total. It therefore supports a test-compilation obstacle in this checkout, not a conclusion that document viewing, editing or conversion is broken.
The repository measured 41,589 files, approximately 1,548,360 lines and 5,619.7 MB. The record detected 24 CI workflows, a Dockerfile, a Compose file and a tests directory. These are repository observations, not proof that those workflows or deployment paths passed. The sandbox configuration was 3 CPUs, 12 GB memory and image lab-rust:1. No new execution, document workload, installed-CLI check or security audit forms part of this review.
Check fidelity with a defined reference
The historical renderer reports should be read with their resolution status. Issue #6095 described a nested-table placement problem with roughly 303 pixels of excess space. It was closed as completed on August 26, 2026, with a maintainer reporting integration of the fix. It should not be presented as an unresolved defect in the measured October checkout. The upstream validation reported there is separate from our run.
A closed issue also does not establish fidelity for every document. For an adoption trial, choose representative files, record the reference application's version, fonts and export route, then compare page count, table continuation, text placement and saved output. The README itself cautions that Hancom PDF output varies by tool, version and output path. Define the reference environment before interpreting a difference.
How to read the scores
Scores are editorial judgments on a five-point scale: 1 means the supplied evidence shows a major obstacle; 2 means substantial constraints; 3 means a usable evaluation starting point with material unknowns; 4 means strong evidence for the stated scope with limited gaps; 5 requires broad corroboration. They are not benchmark results, adoption measurements or compatibility guarantees.
Setup receives 3 because dependency download and build succeeded, while test compilation stopped and the source tree was large. Documentation receives 4 for the supplied English feature and setup documentation, qualified by the Korean source-of-truth roadmap and Korean project records. Community receives 3 for visible contributor and maintainer participation in the supplied issue history; support responsiveness and adoption were not measured. Maturity receives 3 because the documented interfaces and resolved renderer work provide a basis for evaluation, while our run leaves runtime behavior and document fidelity unverified.
For a narrower requirement, hwp.js describes itself as a web HWP viewer and parser library. pyhwp documents HWP v5 stream analysis and extraction, with experimental OpenDocument and plain-text conversion. Neither alternative was run here. Choose based on the operation you need, then verify that operation with your files.

