dat-boris

Context Management Comparison: Opencode vs. Pi

This report compares the context management strategies of the Opencode agent and the Pi agent, based on trace analysis from the TODO app experiment on 2026-04-22 using the Kimi 2.5 model.

1. Overview of Context Management Strategies

Opencode

Opencode employs a highly structured, verbose, and comprehensive system prompt. It explicitly defines:

• Tool Use Protocol: Detailed instructions on how to use tools, the importance of parallel tool calls, and the expected behaviors after tool results (e.g., "If you anticipate making multiple non-interfering tool calls, you are HIGHLY RECOMMENDED to make them in parallel to significantly improve efficiency.").
• Coding Guidelines: Rigid rules for coding from scratch, bug fixes, features, and refactoring (e.g., "Make MINIMAL changes to achieve the goal. This is very important to your performance.").

dat-boris

name	bigquery-schema
description	Fetch and inspect BigQuery table schemas using the bq CLI (part of the Google Cloud SDK). Use when you need to look up column names, types, or structure of a BigQuery table before writing SQL queries or building pipeline assets.

BigQuery Schema Lookup

Use the bq CLI (bundled with gcloud SDK) to inspect table schemas before writing queries or pipeline code.

Fetch a table schema

dat-boris

#!/usr/bin/env python3

"""Given a queue name, export the list of traces which have been annotated to a
CSV file.

./export_langfuse_comments.py <queueId>  <output-csv-file>

The CSV file should contains the following columns:

- query

dat-boris

"""
So, you can give instruction to structured output (SO) in 3 format:
1. Include in desciription in the field of SO
2. Include the JSON schema in prompt.
3. Include it directly in prompt in Natural langauge

If we are given conflicting instructions between prompt and structured output,
which one will win?  This will tell which of the above is the most powerful!

It turns out that:

dat-boris

/** Label Studio include for MathJax.
 *
 * https://docs.humansignal.com/guide/scripts
 *
 * Script src: https://gist.github.com/dat-boris/cc36f2f92fb29d7cb0b00dcabf2d89b6
 */

LSI.import('https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-mml-chtml.js');

function initMathJax() {

dat-boris

# Example URL: https://ai-useast2hub770089131882.services.ai.azure.com/models/chat/completions?api-version=2024-05-01-preview
# - Note the `services.ai.azure.com` domain name, which use `api-key` for auth
# - instead of the `models.ai.azure.com` - which will use `Token bearer`
curl -X POST "${AZURE_SERVICE_URL}" \
    -H "Content-Type: application/json" \
    -H "api-key: ${AZURE_SERVICE_KEY}" \
    -d '{
  "messages": [
    {
      "role": "user",

dat-boris

import cv2
import numpy as np

# Define 3D object points
object_points = np.array([
    [0, 0, 0],
    [1, 0, 0],
    [1, 1, 0],
    [0, 1, 0]
], dtype=np.float32)

dat-boris

#!/usr/bin/env python3
"""A simple script to capture screen, that then post to the server as websocket
for setting up the messages.

We log these folders to allow for the case for following through the data.  The
folder structure works with:

- logs/video/<yyyymmdd-hhmmss>.avi
- logs/keyframes/<yyyymmdd-hhmmss>_<frame>.png

dat-boris

#!/usr/bin/env python
"""A script for downloading video from https://publicorderemergencycommission.ca

Oh my, the site is awful! It streams the mp4 in .ts format that is much slower
than the stream reate.  So this script downloads the stream and compress it using
ffmpeg h264 codec.

You will need to find the video link - you can do that by using the developer tool
when playing the video on page

dat-boris

#!/usr/bin/env python
"""Ensure that you write out loud when you write your text.

    brew install portaudio
    pip installl pynput speechrecognition pyaudio pyttsx3
    ./out_loud.py

WARNING: you might need to hack your security setting following:
https://stackoverflow.com/questions/53088995/pynput-keyboard-listener-does-not-detect-keys-on-mac-os-x
"""