v2.2.0 (feat(worker): record severity in analysis mode, #413) made every
finding carry a severity, and min_severity now applies in both modes
(prompt-manager.ts applies the severity filter unconditionally; only
min_confidence is gated to analysis-only runs).
configuration.md still described the pre-2.2.0 model: min_severity as
exploit-only, findings carrying one rating or the other but never both,
and analysis runs producing no severity at all.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Lead the prerequisites with provider breadth and BYOK instead of Claude.
Add a Continuous Integration section after Quick Start with a working
GitHub Actions example, plus docs/ci-cd.md covering headless execution,
artifact paths, SARIF upload, merge-gating options, and runtime/cost
guidance.
Add three Key Capabilities bullets (machine-readable output, headless
CI/CD execution, BYOK/provider-agnostic) and a question-shaped Common
Questions section before Safety.
Regenerate llms-full.txt, which had drifted from docs/ai-providers.md.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
