Conversation
feat: dialectic debate narrative in PR review output
fix: remove process.env credential pollution re-introduced by merge
refactor: centralize model, file classification, and strategy constants
- Update model references from claude-sonnet-4-20250514 to claude-opus-4-6 - Update workflow example from @v1 to @main - Add review output format section with 3-step debate example - Update Hawk/Owl section to describe STEP 1/2/3 structure - Rewrite CHANGELOG with [Unreleased] section and accurate 1.0.0 stats - Remove outdated Extended Thinking references - Fix test count (104), module count (30), line count (~5,800) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Set language default to "en" (was undefined)
- Make language field non-optional in DialecticConfig
- Fix JSON parser to detect unfenced consensus JSON blocks
When Claude omits code fences, the parser now finds JSON by
matching {"consensus_completed": ...} pattern as fallback,
fixing FP Rate showing "—" and verdict defaulting to LGTM
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
🤖 Dialectic PR ReviewFramework: vanilla SummaryThe language default change ( Issues🐛 Greedy regex with non-greedy inner quantifier may truncate or mismatch JSONFile: The regex Suggestion: Instead of regex-matching nested JSON, use a brace-counting parser or find the outermost 🐛 Non-null assertion on
|
No description provided.