v1.0
Feature release
- Add optional AI-powered analysis with three providers: Claude (Anthropic, default), Gemini (Google AI Studio), local (Ollama vision)
- Add
[ai]extras so SDKs are opt-in and zero-dep installs are unaffected - Per-module HTML reports with embedded chart PNGs and rendered LLM markdown
- Top-level
index.htmlcross-module synthesis when runninganalyze --ai - Three-tier verdict scale:
[HEALTHY]/[WARNING]/[CRITICAL]with concrete remediation suggestions - Deterministic rules engine with severity floor: LLM verdicts can't be downgraded below the worst rule finding
- Configuration-aware prompts: relevant
pg_settingsinjected per module - DBA calibration block in the system prompt to debunk common myths (25% shared_buffers,
random_page_cost=4, etc.) <user_data>envelope around all user-derived data to contain prompt injection
- Add
Maintenance
- Catch SystemExit per module in
analyzeso one module's missing data doesn't kill the whole run - Fix
test_calc_ioratesexpected values for block-size multiplication - Bump Pillow to 12.2.0 for security fixes