|
|
@@ -1,84 +0,0 @@
|
|
|
-# Changelog
|
|
|
-
|
|
|
-## 0.3.6
|
|
|
-
|
|
|
-### Features
|
|
|
-- **10 quality & automation enhancements**
|
|
|
- - Sensitive word detection (dim 27): built-in political/sexual/violence word lists
|
|
|
- - Multi-model routing: `modelOverrides` config maps agent roles to different models
|
|
|
- - Spot-fix revise mode: targeted minimal changes with half token budget
|
|
|
- - Character dialogue fingerprinting: extracts per-character speaking style from recent chapters
|
|
|
- - Entity-based context retrieval: lightweight relevant-summary matching from chapter summaries
|
|
|
- - Post-write constraint validation: regex-based prohibition checks (zero LLM cost)
|
|
|
- - Temperature retry threading: scheduler→runner→writer for quality gate retries
|
|
|
- - Failure clustering: per-dimension failure tracking with diagnostic webhook alerts
|
|
|
- - Auto-detection loop in daemon: audit pass → detect → anti-detect rewrite
|
|
|
- - Analytics CLI: `inkos analytics [book-id]` for chapter/word/audit statistics
|
|
|
-- Word count control via `--words` flag
|
|
|
-
|
|
|
-### Fixes
|
|
|
-- Align capability declarations with actual implementation (genre enum, audit dimensions, truth file count, revise modes)
|
|
|
-
|
|
|
-## 0.3.5
|
|
|
-
|
|
|
-### Features
|
|
|
-- **12-feature enhancement** (Phases 1–12 from plan)
|
|
|
- - Chapter summaries for cross-chapter coherence
|
|
|
- - Structural AI-tells detection (paragraph uniformity, cliche density, formulaic transitions)
|
|
|
- - Webhook notifications with HMAC signing
|
|
|
- - Subplot progress board
|
|
|
- - Emotional arc tracking
|
|
|
- - Character interaction matrix with information boundaries
|
|
|
- - Pacing health scoring
|
|
|
- - AIGC detection API integration (`inkos detect`)
|
|
|
- - Style fingerprint learning (`inkos style analyze/import`)
|
|
|
- - Anti-detect rewrite pipeline stage
|
|
|
- - Smart scheduler with quality gates
|
|
|
- - Detection feedback loop with history analytics
|
|
|
-
|
|
|
-### Fixes
|
|
|
-- Shared prepack/postpack scripts prevent `workspace:*` leaking into npm tarballs
|
|
|
-
|
|
|
-## 0.3.4
|
|
|
-
|
|
|
-### Features
|
|
|
-- CLI UX improvements: auto-detect book-id, `--json` output everywhere, `inkos update` command
|
|
|
-- Variadic args for `review approve/reject`
|
|
|
-
|
|
|
-## 0.3.2
|
|
|
-
|
|
|
-### Features
|
|
|
-- Dual API format support (OpenAI + Anthropic SDK)
|
|
|
-- API key security: key stored in `.env` only, not in `inkos.json`
|
|
|
-- Optional LLM params (temperature, maxTokens)
|
|
|
-- Global LLM config via `~/.inkos/.env`
|
|
|
-- `inkos init` works without arguments (uses current directory)
|
|
|
-- Friendly LLM error messages + `inkos doctor` API connectivity test
|
|
|
-
|
|
|
-## 0.3.0
|
|
|
-
|
|
|
-### Features
|
|
|
-- Three-layer rule architecture: platform → genre → book
|
|
|
-- Genre profiles: xuanhuan, xianxia, urban, horror
|
|
|
-- Genre-aware architect, 19-dimension auditor, reviser modes (polish/rewrite/rework)
|
|
|
-- Genre CLI commands: `inkos genre list/show/edit`
|
|
|
-- Writer prompts extracted into genre-aware prompt builder
|
|
|
-
|
|
|
-## 0.2.0
|
|
|
-
|
|
|
-### Features
|
|
|
-- Pluggable radar via `RadarSource` interface
|
|
|
-- State snapshots and chapter rewrite command
|
|
|
-- Numeric verification rules
|
|
|
-- Batch approve command
|
|
|
-- Export command for publishing
|
|
|
-- Book write lock with index-based chapter numbering
|
|
|
-
|
|
|
-## 0.1.0
|
|
|
-
|
|
|
-### Initial Release
|
|
|
-- Multi-agent pipeline: Writer → Auditor → Reviser
|
|
|
-- CLI with init, config, book, write, review, status, radar, up/down commands
|
|
|
-- Three truth files: current_state, particle_ledger, pending_hooks
|
|
|
-- Architect agent for world-building and volume outlines
|
|
|
-- OpenAI-compatible LLM provider
|