v0.20.2

Pitloom 0.20.2 is a patch release for AI models. It reads CRFsuite models (labels, counts and a label-based description) and reads ONNX models' name, version and licence. Model annotations now carry collections as arrays, scalars as text with valueTypes, and embedded JSON follows RFC 8785. Over-long model names are cut with a digest and each cut gets a WARNING:, and invisible or bidi characters in model text are shown as \uXXXX. It requires licenseid 0.4.3.

Note

Known issue: a CRFsuite model's generated description (built from its labels) can replace a description from a fragment or catalogue, with a WARNING:. A fix is planned for 0.21.0.

Install for CLI/Python API:

pip install -U pitloom

Install for AI coding assistant:

See the website on how to install Pitloom AI skills and the Claude Code plugin.
Try asking the AI agent to "generate SBOM for this project".

Report issues: https://github.com/bact/pitloom/issues

What's changed

• Name AI models by file; read ONNX licence, version by @bact in #292
• Speed up CI: worksteal, uv installs, separate pylint job by @bact in #295
• Windows tests: TEMP on D:, upload per-test timings by @bact in #297
• Add CRFsuite support; uniform, safer model metadata by @bact in #294

Full Changelog: v0.20.1...v0.20.2