I’ve used AI every day for the past 3 years. Claude, ChatGPT, n8n, Perplexity...
I’ve automated almost everything:
- write content 10 times faster
- handle email in 5 minutes instead of 2 hours
- do the work of 3 people in a single day
I’ve used AI every day for the past 3 years. Claude, ChatGPT, n8n, Perplexity...
I’ve automated almost everything:
Status: Draft v1 (language-agnostic)
Purpose: Define a service that orchestrates coding agents to get project work done.
Symphony is a long-running automation service that continuously reads work from an issue tracker (Linear in this specification version), creates an isolated workspace for each issue, and runs a
| #!/bin/bash | |
| echo "=== Axios Compromise Scanner ===" && echo "Malicious: 1.14.1, 0.30.4 | Attack: 2026-03-31" && echo && find ~ -path "*/node_modules/axios/package.json" -type f 2>/dev/null | while read p; do v=$(jq -r '.version' "$p" 2>/dev/null || grep -o '"version"[[:space:]]*:[[:space:]]*"[^"]*"' "$p" 2>/dev/null | head -1 | cut -d'"' -f4); case "$v" in 1.14.1|0.30.4) echo "🚨 MALICIOUS: $p → $v" ;; *) echo "✓ Safe: $p → $v" ;; esac; done && echo && if find ~ -path "*/node_modules/plain-crypto-js" -type d 2>/dev/null | head -1 | grep -q .; then echo "🚨 TROJAN: plain-crypto-js present"; else echo "✓ No trojan package"; fi && echo && if find ~ -path "*/node_modules/axios/*" -type f -name "*.js" -exec grep -l "sfrclak\.com" {} \; 2>/dev/null | head -1 | grep -q .; then echo "🚨 C2 CALLBACK: sfrclak.com found"; else echo "✓ No C2 callbacks"; fi |
| #!/bin/bash | |
| echo "=== Axios Compromise Scanner ===" | |
| echo "Malicious versions: 1.14.1, 0.30.4" | |
| echo "Checking..." | |
| echo | |
| find ~ -path "*/node_modules/axios/package.json" -type f 2>/dev/null | while read p; do | |
| v=$(jq -r .version "$p" 2>/dev/null || grep -oP '"version":\s*"\K[^"]+' "$p" | head -1) | |
| case "$v" in | |
| 1.14.1|0.30.4) echo "🚨 MALICIOUS: $(dirname $p) → $v" ;; |
| # Axios NPM Compromise Scanner | |
| # Malicious versions: 1.14.1, 0.30.4 | Attack: March 31, 2026 | |
| # https://www.stepsecurity.io/blog/axios-compromised-on-npm-malicious-versions-drop-remote-access-trojan | |
| ## One-liner (curl | bash) | |
| ```bash | |
| curl -sL https://gist.githubusercontent.com/tuannvm/bceae613979f7b91586c75262ca37b81/raw/axios-scan.sh | bash | |
| ``` | |
| ## Direct one-liner |
| #!/usr/bin/env bash | |
| set -u | |
| CCODEX_CMD="${HOME}/.local/bin/ccodex" | |
| ALIAS_FILE="${HOME}/.oh-my-zsh/custom/ccodex-alias.zsh" | |
| CCODEX_INSTALL_URL="${CCODEX_INSTALL_URL:-https://gist.github.com/tuannvm/6fe6cf37c05265a9ee0acbd3d2da52d4/raw/ccodex}" | |
| has_cmd() { | |
| command -v "$1" >/dev/null 2>&1 |
https://blog.cloudflare.com/code-mode/
Cloudflare’s blog post “Code Mode: the better way to use MCP,” authored by Kenton Varda and Sunil Pai, introduces a new approach to using the Model Context Protocol (MCP) that significantly improves AI agent performance. Traditional MCP usage involves directly exposing tools to large language models (LLMs), but this method faces limitations due to LLMs’ limited familiarity with tool-call tokens, resulting in difficulty with complex or numerous tools.
The new Code Mode approach converts MCP tools into a TypeScript API and instructs the LLM to write code that calls the API. This provides several advantages:
argocd cluster add docker-desktophttps://registry.terraform.io/providers/oboukili/argocd/latest/docs/resources/cluster
| package main | |
| import ( | |
| "fmt" | |
| "io/ioutil" | |
| "log" | |
| "net/http" | |
| "github.com/gorilla/mux" | |
| ) |