v0.1.0

Summary

• Hardened trigger-extraction logic: Windowed multi-token matching, merged duplicate scoringCues for the same trigger at load-time, and added warnings for cross-trigger overlaps.
• Updated config/trigger-extraction.json: Collapsed literal duplicate scoringCues blocks and tightened the non-consensual pattern generation regex so generate <number> requires deceptive/mass-generation context (e.g., "generate N phishing emails").
• Improved scripts/normalize-percases.ts: Unified repoRoot resolution (defaultRepoRoot()), added --apply-signal-ids behavior to recompute signal_ids from extracted triggers, explicit dry-run messages, and clearer error surfacing.
• Added and strengthened tests:
  • scripts/__tests__/extract-triggers.test.ts - negative and positive cases for generate N and other extraction rules.
  • scripts/__tests__/normalize-percases.unit.test.ts - runs extraction + --apply-signal-ids, verifies old SIDs are replaced and mapped/derived SIDs are present.
  • golden/regression tests added for extractor outputs.
• Misc: Refactors and small fixes to keep deterministic ordering and canonical mapping; CI/tests pass locally.

Notable Commits

• a71ec2d: fix: harden trigger extraction and signal ID synchronization (#33)
• 5693106: feat: add rule-based trigger extraction for scenarios (#30)
• f1adb80: CI: keep automatic triggers; disable image generation by default
• 1713e82: feat(confidence): normalize confidence scoring and retune trigger weights (#18)

Testing

• Full test suite passed locally: npm test -- --runInBand.

Notes

• The extractor will log warnings for phrases/cues that appear in multiple triggers; these are cross-trigger overlaps that may need manual semantic resolution in future config edits.

Architecture Pipeline