The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The README.md file recommends installing the Kiro CLI tool using a high-risk remote execution pattern.
Evidence: curl -fsSL https://cli.kiro.dev/install | bash in README.md.
[PROMPT_INJECTION]: The skill provides instructions that explicitly bypass the safety protocols and user confirmation prompts of the underlying AI agents.
Evidence: The skill instructs the agent to use --trust-all-tools for Kiro CLI to skip confirmation prompts and --yolo for Codex, which is described as having "NO sandbox, NO approvals (fastest, most dangerous)" in SKILL.md.
The agent is encouraged to use mktemp -d && git init to bypass Codex's restriction against running in untrusted directories.
[COMMAND_EXECUTION]: The skill relies on the bash tool to execute arbitrary shell commands for launching various third-party coding agents.
Evidence: Multiple instances of bash pty:true command:"..." across SKILL.md for running kiro-cli, claude, codex, opencode, and pi.
[EXTERNAL_DOWNLOADS]: The skill references and installs external software from third-party repositories and domains.
Evidence: Downloads from https://cli.kiro.dev/install and installation of the Node.js package @mariozechner/pi-coding-agent.
[DATA_EXFILTRATION]: While no direct exfiltration is present, the coding agents are granted broad read/write access to the user's filesystem (workdir), which creates a significant risk if the agent is influenced by indirect prompt injection from analyzed code.

coding-agent