large-repo-git

---

name: large-repo-git description: Safe, fast git operations for large repositories that time out on naive commands. Covers: status, fetch, diff, log, sync-from-remote, find-files, and avoid-kills. Use when any git operation is slow, killed, or timing out. Aliases - bigrepo, git-fast, repo-sync, git-safe, large-repo

large-repo-git — Fast git for large repos

This repo is large (~18MB dashboard JSON, 200+ strategy files, coinglass DB, binary assets). Naive commands like git status -uall, find /mnt/c/..., and git fetch without --depth will time out or get killed. This skill is the safe replacement for every common git operation.

---

Rule #1 — Never use these commands on this repo

❌ Dangerous (will kill/timeout)	✅ Safe replacement
git status -uall	git status --short (no -u flag)
find /mnt/c/... -type f	git ls-files --others --exclude-standard | head -20
find . -name "*.py"	git ls-files "*.py" | head -30
git diff (full)	git diff --stat then git diff -- <specific_file>
git log (no limit)	git log --oneline -10
git fetch (full)	git fetch --depth=1 origin main
git clone (full)	git clone --depth=1 --filter=blob:none
grep -r pattern .	git grep pattern (respects .gitignore, faster)
wc -l on large files	head -5 file + git show --stat HEAD:file
git diff HEAD~50	git diff HEAD~5 --stat (always bound the range)

---

Rule #2 — WSL path discipline

When running from WSL (/mnt/c/... paths), the Windows filesystem is 10-30× slower than native Linux paths. Symptoms: Task killed in 1m26s, Task killed in 2m0s.

# BAD — WSL accessing Windows FS, slow
cd /mnt/c/findtorontoevents_antigravity.ca && git status

# GOOD — same thing, explicit timeout awareness
cd /mnt/c/findtorontoevents_antigravity.ca && timeout 20 git status --short

If you repeatedly hit kills, the agent should be running from the Windows side (PowerShell / cmd), not WSL. Native Windows git is faster on this repo.

---

Status (safe)

# What's modified (tracked files only — fast)
git status --short

# Staged vs unstaged counts only
git diff --name-only | wc -l
git diff --cached --name-only | wc -l

# Untracked sample (first 10 only — never drain all)
git ls-files --others --exclude-standard | head -10

---

Fetch / Pull (safe)

# Shallow fetch — gets latest without full history download
git fetch --depth=1 origin main

# Fast-forward pull (no rebase drama, no merge commit)
git pull --ff-only origin main

# Full safe sync sequence (stash → pull → pop)
git stash && git pull --rebase origin main && git stash pop

# Check how far behind we are WITHOUT fetching
git rev-list --count HEAD..origin/main   # requires prior fetch

---

Diff (safe)

# What changed — stats only first
git diff --stat
git diff --cached --stat

# Then drill into one file
git diff -- path/to/specific_file.py

# Compare to origin without full fetch
git diff HEAD origin/main --stat   # after fetch --depth=1

---

Log (safe)

# Recent commits
git log --oneline -10

# Since a specific date (avoids full history traversal)
git log --oneline --since="2 days ago"

# Changes in one file
git log --oneline -5 -- audit_trail/quality_gates.py

# Commits ahead of origin
git log --oneline origin/main..HEAD

---

Find files (safe)

# Find tracked Python files (fast — uses git index)
git ls-files "*.py" | head -30

# Find tracked files containing a pattern (fast — uses git index)
git grep -l "BLOCKED_ASSET_STRATEGY_PAIRS" | head -20

# Find recently modified tracked files
git diff --name-only HEAD~5

# Find untracked files (sample only)
git ls-files --others --exclude-standard | head -20

---

Sync from remote (full workflow)

Use this instead of any ad-hoc fetch+pull sequence:

# 1. Check if we need to pull at all
git fetch --depth=1 origin main
git rev-list --count HEAD..origin/main   # 0 = already up to date

# 2. If behind, safe pull
git stash
git pull --rebase origin main
git stash pop

# 3. Verify
git log --oneline -3

---

Push (safe)

# Always unset stale GITHUB_TOKEN first (this repo has a known expired PAT in env)
unset GITHUB_TOKEN

# Push
git push origin main

# If rejected (remote has new commits):
git stash && git pull --rebase origin main && git stash pop && git push origin main

Known issue on this repo: GITHUB_TOKEN env var may contain an expired PAT (github_pat_11AJHZILQ...). Always unset GITHUB_TOKEN before push. If gh auth status shows the env var token, run Remove-Item Env:GITHUB_TOKEN in PowerShell (or unset GITHUB_TOKEN in bash) to fall back to the keyring token.

---

Commit (safe)

# Add specific files only — never `git add -A` on this repo
# (coinglass.db, dashboard_data.json, binary assets will bloat the commit)
git add path/to/specific_file.py path/to/another_file.md

# Commit with heredoc (avoids quoting issues)
git commit -m "$(cat <<'EOF'
feat(scope): what changed and why

Co-Authored-By: Claude Sonnet 4.6 <[e-mail]>
EOF
)"

---

Large file traps to avoid

File	Size	Why dangerous
audit_dashboard/data/dashboard_data.json	~18MB	Never cat — use python -c "import json; d=json.load(open(...))" with targeted key access
coinglass_strategies/data/coinglass.db	~27MB	Binary — never diff, never cat
audit_dashboard/index.html	~4MB	Auto-generated — never edit
logs/cross_pc_protocol/events.jsonl	Growing	Tail only: tail -20 logs/...

Read large JSON files with targeted key access:

python -c "
import json
from pathlib import Path
d = json.loads(Path('audit_dashboard/data/dashboard_data.json').read_text(encoding='utf-8'))
# Access only what you need:
print(d.get('generated_at'))
print(list(d.get('performance', {}).get('asset_class_health', {}).keys()))
"

---

When commands are killed — diagnosis

If a command is killed (Task killed in Xs), work through this checklist:

1. Add timeout N — wrap command: timeout 15 git status --short
2. Check if running in WSL on Windows FS — if path is /mnt/c/..., move to PowerShell
3. Scope the command — add | head -N or -- specific_file to limit output
4. Avoid find — use git ls-files instead
5. Avoid unscoped git diff — always add --stat first, then drill into one file
6. Check if large file is being read — check if the killed command touches the 18MB dashboard JSON or 27MB DB

---

Companion skills

• /check-gh-actions — verify CI after push
• /cross-pc-health — gateway pre-flight
• /large-repo-read — safe read patterns for large JSON/log files (see below)