v2.0.0

2.0.0 (2026-10-07)

⚠ BREAKING CHANGES

• ship the skill from skills/better-writing only (#62)

Bug Fixes

• ship the skill from skills/better-writing only (#62) (3e23468)

v1.3.0

1.3.0 (2026-09-29)

Features

• add a trigger eval for the skill description (#50) (6efb0a5), closes #41
• evals: add drafting and review fixtures (#51) (7455f66)
• evals: add fixtures for venue rules, UI text and reference docs (#49) (cd3f9ad)
• evals: commit a reference summary and diff runs against it (#53) (bd5ce9f)
• evals: grow the held-out split to 9 of 22 fixtures (#52) (e5bddb3)

v1.2.1

1.2.1 (2026-09-26)

Bug Fixes

• keep the detector caveat out of the text, add two eval fixtures (#28) (731d4ff)

v1.2.0

1.2.0 (2026-09-26)

Features

• fold in September 2026 detection and editing research (#24) (65f15e6)

v1.1.1

1.1.1 (2026-09-25)

Bug Fixes

• stop the skill inventing content, isolate eval runs, tighten checks (#19) (f79b604)

v1.1.0

1.1.0 (2026-09-25)

Features

• install as a Claude Code plugin, add before/after GIF (#17) (2ee1a1d)

v1.0.0

1.0.0 (2026-09-22)

Features

• add evals, before/after examples, and differentiation-focused docs (aca2611)
• add model-backed eval runner, harden checker, align docs with the catalogue (33d998d)
• add patterns from the 2026 skills survey and Wikipedia cleanup work (cc2ab77)
• add unslop-derived patterns to the catalogue (0f6ea78)
• contrast checks, no-skill baseline, and pairwise comparison in the evals (6900fde)
• detection research refresh (2026-09-22) (#7) (8f40091)
• fiction genre bank, chat register, and in-place file editing rules (1ebc497)
• give voice a procedure and measurable checks (bedbfe8)
• init (0cdbb04)
• refresh AI-writing catalogue with tiers, evidence, and genre tells (061fe77)
• voice-drift metrics for keep-my-voice fixtures (8a55285)

Bug Fixes

• a kept hope stays a hope, with nothing wrapped around it (#10) (5fbb992)
• address deepseek and coderabbit review feedback (1248687)
• align dash policy, exempt range en dashes, repair validation docs (5aae92f)
• align placeholder rule, repair cross-ref, add CI and repo validator (4023076)
• apply verified review findings across skill, references, evals, and CI (d943bb1)
• bump agents/openai.yaml as a generic file so release-please keeps its comments (#16) (7bf2fad)
• close checker gaps and consistency issues from repo review (#6) (6c3b860)
• correct catalogue entries the 2025 to 2026 research contradicts (60332b7)
• drop faceless-authority nouns without inventing a source (#9) (8d7de15)
• forbid additions in rewrites and add an added-claims judge to the eval runner (7baeaf0)
• judge sees the brief, catches hope-to-fact rewrites, survives bad replies (#12) (cb49153)
• keep dash rates diagnostic-only and stop concretising vague sources (dd9d0a5)
• light-edit mode and no invented endings (#8) (a18a64b)