Automate a weekly report

Goal: have PiG write the same report from fresh data every week, without you typing anything. You do the task once by hand, save the instruction as a small script and schedule the script.

You need: PiG signed in with pig login (Your first ten minutes), a data file that changes weekly (CSV or text) and a Mac or Linux computer. The script is a shell script. On Windows, run it with WSL or Git Bash.

Time: fifteen minutes to set up. Each weekly run takes as long as one model reply.

How it works

pig -p "<prompt>" runs one prompt and exits, with no screen to watch and nothing to type. That is print mode. The pieces of the command are:

Step 1: do it once by hand

Make a folder with the weekly data and start PiG there:

mkdir ~/weekly && cd ~/weekly
printf 'day,orders,refunds\nMon,41,2\nTue,38,1\nWed,52,4\nThu,47,0\nFri,60,3\n' > this-week.csv
pig

Read this-week.csv. Write weekly-report.md with three lines: total orders, total refunds with the refund rate as a percentage, and the best day by orders.

Check weekly-report.md against the numbers. Here the orders total 238 and refunds total 10, which is 4.2 percent, and the best day is Friday. Adjust the prompt until the report is what you want. The final prompt is what you will schedule, so write it so that a stranger could follow it.

Step 2: save it as a script

Save this as ~/weekly/weekly-report.sh. Replace /usr/local/bin/pig with the output of command -v pig:

#!/bin/sh
# Writes weekly-report.md from this-week.csv. Appends everything PiG prints to weekly-report.log.
cd "$(dirname "$0")" || exit 1
/usr/local/bin/pig -p --no-session --no-extensions --tools read,write @this-week.csv \
  "Write weekly-report.md: total orders, total refunds, refund rate and the best day. Three lines." \
  >> weekly-report.log 2>&1

Make it runnable and run it:

chmod +x ~/weekly/weekly-report.sh
~/weekly/weekly-report.sh

PiG prints the model's final reply into weekly-report.log and writes weekly-report.md:

Report written to weekly-report.md.

# Weekly report

Orders: 238. Refunds: 10 (4.2%). Best day: Friday (60 orders).

The script's exit code is PiG's exit code. It is 0 when the run worked. A model that cannot be reached writes its error to the log and the script exits 1. Here is the whole log of a run when the provider could not be reached:

Connection error.

Look in weekly-report.log first when a report is missing.

Why the full path and a stored login

A scheduler starts the script with an almost empty environment. It has no PATH entry for your PiG, and it does not read your shell profile, so a key you exported in ~/.zshrc is not there. Two rules follow:

1. Write the full path to pig, as above.
2. Sign in with pig login. PiG stores that credential in ~/.pig/agent/auth.json and reads it in every run (providers.md: stored credentials take precedence over the environment). If you use an API key from the environment, put the export line in the script itself, above the pig line, and protect the script (chmod 700).

The script was tested under env -i PATH=/usr/bin:/bin HOME=..., an almost empty environment like the scheduler's.

The scheduler's PATH also has no Node.js. PiG cannot load an extension written in TypeScript or JavaScript there, and the run would stop with Error: Failed to load extension ... node was not found on PATH. The report needs no extensions, so the script starts PiG with --no-extensions.

Step 3: schedule it

crontab -e opens your schedule. Add one line. This one runs every Monday at 08:00:

0 8 * * 1 /home/you/weekly/weekly-report.sh

The five fields are minute, hour, day of month, month and day of week (1 is Monday). Check the line you saved:

crontab -l

On a Mac, cron may need Full Disk Access (System Settings, Privacy & Security) before it can read folders such as Documents. Test the schedule by setting the line to run in two minutes, then set it back.

Step 4: keep it safe

• Keep the tool list small. --tools read,write is the point of the script. A scheduled run has nobody watching, so give it only the tools it needs. Leave out bash.
• One output file. The prompt names weekly-report.md. write replaces a file of that name each week. Keep last week's with a dated copy: add cp weekly-report.md "report-$(date +%F).md" as a line before the pig line.
• Read the first few reports. A model can misread a column. Spot-check against the data until you trust the prompt.
• No data, no report. A missing @ file stops PiG before any model call: it prints Error: File not found: <path> and exits 1, and the log shows it (Pi packages/coding-agent/src/cli/file-processor.ts:36-40). An empty file is skipped without any message (packages/coding-agent/src/cli/file-processor.ts:46-49), so the model gets your instruction and no data, and may make numbers up. Add [ -s this-week.csv ] || exit 1 after the cd line to stop on both.

Undo

Remove the schedule line (crontab -e, delete the line) and delete weekly-report.sh. The script wrote only weekly-report.md and weekly-report.log, because it has no tool to write anywhere else. It could still write to a path the prompt names, so keep paths in the prompt local.

Variations

• A report from several files: @sales.csv @refunds.csv in the same command.
• A longer instruction: keep it in a file and use @prompt.md in place of the quoted text.
• A different model for this job: add --model provider/model, for example a smaller and cheaper one.