deep-analysis

npx skills add majiayu000/claude-skill-registry --skill deep-analysis-gregorstocks-mage-bench --agent claude-code

Same command for any agent — swap --agent for codex, cursor, copilot.

# Deep Analysis Deep analysis of game logs — reads all raw log files, traces bugs to source code, and files detailed issues. For quick triage, use `/fast-analysis` instead. ## Workflow ### Step 1: Select the game Determine which game to analyze: - If the user specified a game ID (e.g. `game_20260211_080409`), use that. - If the user said "most recent" or similar, find the latest: ```bash uv run python scripts/list_recent_games.py ``` - If the user mentioned a config name (e.g. "round-robin-commander", "jumpstart-dumb", "modern-staller"), use the corresponding symlink: ```bash uv run python scripts/list_recent_games.py --config {config} ``` where `{config}` might be `round-robin-commander`, `jumpstart-dumb`, `modern-staller`, etc. Check what symlinks exist with `--symlinks`. - **If no game specified at all**, find the most recent unanalyzed game: ```bash uv run python scripts/analysis/find_unanalyzed.py --type deep --count 1 ``` Set `GAME_DIR=~/.mage-bench/logs/{game_id}`. If the full log directory doesn't exist but the game export exists in `website/public/games/` (`.json` or `.json.gz`), tell the user the full logs aren't available and offer to run a fast analysis from the export

uv run python scripts/list_recent_games.py
uv run python scripts/list_recent_games.py --config {config}
uv run python scripts/analysis/find_unanalyzed.py --type deep --count 1
uv run python scripts/game_gz_bootstrap.py ${GAME_ID}
uv run python scripts/query_issues.py
git log --oneline --since="YYYY-MM-DD" origin/master  # date of the game