Returns evaluation in centipawns and win%, the best lines in SAN, material
balance, and simple tactical flags (fork threats, pins). Use this to explain
*why* a move is best in a specific position.

Each move gets a classification (best/good/inaccuracy/mistake/blunder with
Korean labels), win% before/after, centipawn loss, the engine's preferred
move, and tactical-motif tags. The summary gives per-side accuracy/ACPL,
per-phase breakdown, and the worst moments (with FENs for drilling).

Set user_color to 'white' or 'black' to focus the worst-moment and motif
reporting on one player.

source: 'lichess' or 'chesscom'. speed (optional) filters by
bullet/blitz/rapid/classical. Returns normalised game records including PGN,
players, ratings, result and opening. No analysis is run here.

Fetches recent games, analyses each from the player's perspective, and
aggregates: overall accuracy/ACPL, per-phase (opening/middlegame/endgame)
breakdown, recurring tactical blind spots (with example positions), leaky
openings, a time-trouble proxy, and a ranked list of top weaknesses with
example FENs. This is the core 'coach' tool — heavier (runs the engine over
many games), so keep max_games modest for fast turnaround.

Runs diagnose_weaknesses, then returns re-solvable drill positions taken
from the player's actual blunders (FEN, side to move, the engine's preferred
move as the answer), ordered to target their top weaknesses, plus an
optional Lichess daily puzzle as a warm-up.