Release v0.7.0 — Opening & Training Tools · rutvij26/chess-context

What's new

get_opening_theory — fetch key ideas, main continuations, and win stats for any opening by FEN or name (via Lichess Explorer)
find_opening_gaps — replay your stored games to find positions where you're repeatedly surprised by opponent deviations
generate_puzzles — extract tactical puzzles from your own games, filtered by difficulty and type

JSONB double-encoding: insertAnalysis used JSON.stringify() which postgres.js re-serialized as a JSONB string instead of an array — move_records and critical_moments were silently returning [] for every game. Fixed with db.json().
NUMERIC coercion: postgres.js returns NUMERIC(5,2) as strings ("0.00"), breaking white_accuracy === 0 comparisons. Fixed at the store layer.
Identical style fingerprints: all users got the same scores because the above two bugs left critical_moments and move_records empty for every analysis.
Tactical sharpness formula: looked for eval_drop_cp < 30 but the pipeline only stores drops ≥ 50 — found was always 0. Threshold changed to < 200 (not a blunder = success).
Aggression multiplier: was × 20, producing scores of 2–10 that never reached the 40/70 label thresholds. Changed to × 200.
Style label routing: added Tactical Fighter and Cautious Strategist — previously everyone with low aggression fell through to "Solid Positional Player".
Result format mismatch: scoreEndgameSkill and detectEndgameTechniqueIssue compared against "1-0"/"0-1" but the DB stores "win"/"loss"/"draw".
Auto re-queue: refresh_games now automatically re-queues analyses that ran before Stockfish was warm (zero accuracy scores).

Add Contributor Covenant Code of Conduct by @rutvij26 in #74
feat: add get_opening_theory, find_opening_gaps, generate_puzzles (v0.7.0) by @rutvij26 in #75
fix: JSONB double-encoding, identical style fingerprints, and broken style scoring by @rutvij26 in #77

Full Changelog: v0.6.0...v0.7.0