Changelog

Features

• learn: mine rules with 4 concurrent lenses, self-repair, live progress (#73) (4200417)

Bug Fixes

• judge: bump axllm to 23.0.3 to fix Anthropic opus-4-8 400s (#71) (3314938)

Bug Fixes

• ci: do not fail the commit-message check on a not-yet-created release tag (#69) (7634a9f)

⚠ BREAKING CHANGES

• local: model specs no longer run. Replace them with openai:\<base-url\>#\<model\> pointed at a local OpenAI-compatible server, or with a hosted spec. The --acknowledge-local flag and the ai.localAcknowledged config field are removed; a stale field is ignored rather than rejected.

Features

• harvest no-ai-slop writing rules (#63) (dd33aed)
• judge: per-rule request planning, parallel judging, configurable token cap (#64) (c69d779)
• remove in-process local model inference (#67) (6270470)

Bug Fixes

• docx: write the document when findings are comment-only (#68) (df0aac6)

Features

• ai: run AI rules by default whenever a model and credentials resolve; --no-ai opts out (#59) (cfeea0b)
• config: user-level config moves to ~/.lawlint/, with project config layering over it field by field (#59) (cfeea0b)
• docx: review copies comment on every finding, not only fixable ones; anchors may overlap and span runs (#59) (cfeea0b)
• output: default to a per-tier coverage summary; add --list, --coverage, and --format full (#59) (cfeea0b)
• rules: fixTemplate: capture-group interpolation for phrase rules (#59) (cfeea0b)
• rules: hard/soft rule terminology + unified Markdown skill-file rule format (#55) (840c7e3)
• website: adopt the Blume framework, docs at /docs (6de5972)

Bug Fixes

• fix: apply MaybeIncorrect judge fixes instead of filtering both fix paths to MachineApplicable (#59) (cfeea0b)

Features

• cli: TUI-native setup wizard that launches lawlint (#53) (004e48e)

Features

• ai: cloud-first defaults — hosted providers lead, local is acknowledged opt-in (#50) (#51) (6009a7c)
• cli: lawlint learn — mine a personal rule package from the user’s corpus (#40) (#46) (34c844e)
• core: layer-2 statistical rules — burstiness and triad density (#37) (#48) (db3e0ee)
• core: split rule intent (style|detection); score aggregates detection rules only (#38) (#43) (9831e26)
• docx: lint and fix .docx via tracked changes + comments (#34) (f3c2255)
• eval: judged evaluation — tier-3 metrics and verdict-discipline rate (#39 Part 2) (#49) (d91573b)
• init: AI model preferences — local/hosted catalog, keys outside the project (#41) (#44) (3e7c3d3)
• rules: add Orwell writing rules and AI-voice checks (#47) (29577be)

Bug Fixes

• judge: drop negative-verdict findings and demand [] for clean chunks (#42) (0a2994f)

Features

• mechanical fixes, diff view, and AI remediation prompts (#32) (75ef160)
• website: simplify navigation, hero, and docs structure (#28) (d8ae603)

Features

• cli: add lawlint init and .lawlint/config.json discovery (#25) (528be79)
• cli: launch ratatui TUI on bare lawlint invocation (#24) (311219e)
• website: add changelog page rendering release notes (#27) (350326e)

Features

• cli: add –version, auto update-check, and self-update (#21) (cd61088)

Features

• rebuild playground as a React island with shadcn/ui + Tailwind (#15) (f9d9965)
• rewrite rules engine as 3-tier declarative system with local AI judge (#17) (f0eed56)
• website: add favicon from desktop app icon (#16) (74fd62f)