| # tac — cat backwards: last line first | |
| # (tac is cat spelled backwards) | |
| tac logfile.log # reverse entire file | |
| tac logfile.log | head -20 # last 20 lines, newest-first | |
| tac logfile.log | grep -m1 'ERROR' # last occurrence of ERROR | |
| # Process a JSONL file newest-first: | |
| tac data.jsonl | head -5 | python3 -c " | |
| import sys, json |
| # comm — line-by-line comparison of two sorted files | |
| # Input must be sorted. Use: sort file | sponge file | |
| comm a.txt b.txt # 3 columns: only-A | only-B | both | |
| comm -23 a.txt b.txt # only in A (suppress cols 2 and 3) | |
| comm -13 a.txt b.txt # only in B (suppress cols 1 and 3) | |
| comm -12 a.txt b.txt # in both (suppress cols 1 and 2) | |
| # Sort on the fly with process substitution: |
| # column — align delimited input into a readable table | |
| # TSV file: | |
| column -t -s $'\t' models.tsv | |
| # CSV: | |
| column -t -s ',' data.csv | |
| # Colon-separated: | |
| column -t -s ':' /etc/passwd |
| # sponge — soak up all stdin, then write (from moreutils) | |
| # sudo apt install moreutils | |
| # THE PROBLEM: this destroys the file before sort reads it | |
| sort file.txt > file.txt # ⚠ race condition — file truncated on open | |
| # THE FIX: | |
| sort file.txt | sponge file.txt # reads all input, then writes atomically | |
| # In-place JSON formatting: |
| # ts — timestamp every line of stdin (from moreutils) | |
| # sudo apt install moreutils | |
| command | ts # May 25 13:01:02 <line> | |
| command | ts '[%H:%M:%S]' # [13:01:02] <line> | |
| command | ts '[%Y-%m-%dT%H:%M:%S]' # [2024-05-25T13:01:02] <line> | |
| command | ts -s '%.s' # relative: seconds since last line | |
| # The full AI logging pattern — timestamp, merge stderr, save + display: | |
| agent-command 2>&1 | ts '[%H:%M:%S]' | tee run.log |
| # pv — add a progress bar to any pipeline | |
| pv input.jsonl > /dev/null # bytes + throughput | |
| pv --line-mode input.jsonl | process-tool # line count mode | |
| # Drop into any pipeline — transparent pass-through: | |
| cat input.jsonl | pv | python3 process.py | |
| # ETA with known total: | |
| LINES=$(wc -l < input.jsonl) |
| # tee — write to stdout AND a file at the same time | |
| <command> | tee output.log # overwrite | |
| <command> | tee -a output.log # append | |
| <command> | tee file1 file2 # fan out to multiple files | |
| # The AI workflow pattern — log + timestamps + live view in one pipeline: | |
| agent-command 2>&1 | ts '[%H:%M:%S]' | tee run-$(date +%Y%m%d-%H%M%S).log | |
| # Process and preserve the original simultaneously: |
| # watch — repeat a command every N seconds with a live-updating view | |
| # Basic: refresh every 2 seconds | |
| watch -n2 <command> | |
| # Highlight what changed since last refresh | |
| watch -n1 -d ls -lh outputs/ | |
| # Monitor GPU memory during inference | |
| watch -n1 nvidia-smi --query-gpu=memory.used,memory.free --format=csv |
Reason through git workflow problems and surface the right advanced git command. Each section maps a real situation to a specific tool with concrete commands to run.
Invoke this skill when the user:
| # git bundle — the git sneakernet | |
| # No network. Air-gapped machine. USB drive. It still works. | |
| # Pack everything — every ref, every object: | |
| git bundle create repo.bundle --all | |
| # Verify the bundle is self-contained: | |
| git bundle verify repo.bundle | |
| # Clone from the bundle (it's a valid git remote): |