pw-discoverylisted
Install: claude install-skill productweaver/productweaver-plugin
# pw-discovery: fold interviews into an ODI knowledge base
*A free skill from [Product Weaver](https://productweaver.com).*
One JSON file is the knowledge base (KB). Each run reads new transcripts, extracts ODI knowledge with verbatim evidence, and merges it into the KB through `scripts/kb_tool.py`. The script checks every quote against its transcript and refuses the whole merge if one does not match. That makes the KB safe to build on over many runs: nothing in it is paraphrased evidence or a duplicate.
## Running the script
The script needs Python 3.9 or newer and nothing else: it uses only Python's standard library, so `kb_tool.py` installs nothing and needs no network. It finds its schemas and the map template next to itself, so run it from the root of the workspace, where the study lives:
```
python3 <skill folder>/scripts/kb_tool.py <command> ...
```
Below, `kb_tool` stands for that line. Do not write plain `python`; many machines do not have it. On Windows, if `python3` is missing, use `py -3`.
Write the script's full path in every command: never put it in a shell variable (`S=<skill folder>`), never `cd` into the skill folder, and never run it in a `for` loop: write each command out. Claude Code cannot check such a command, so it asks the user or refuses it. Several `kb_tool` commands may share one call.
Transcripts can be `.md`, `.txt`, `.docx` or `.pdf`. The script reads `.docx` itself. It reads `.pdf` with pypdf when that package is installed. When it is no