The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill recommends engineering workflows that involve remote code execution. Specifically, in principles/new-project-defaults.md, it provides templates for CLI binary distribution using piped shell commands: curl -sSfL .../install.sh | bash for Unix-like systems and irm .../install.ps1 | iex for Windows. While the instructions advise for checksum verification, the pattern itself is a high-privilege execution vector. Additionally, languages/typescript.md suggests installing the anti-slop utility using npx, which downloads and executes code from a remote source.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to process and refactor source code, creating a surface for indirect prompt injection attacks. * Ingestion points: The agent reads and analyzes source code across multiple languages (Rust, Go, Python, TypeScript) provided in the project context. * Boundary markers: No specific delimiters or safety instructions are defined to separate the skill's instructions from potentially malicious code comments or strings in the analyzed files. * Capability inventory: The skill authorizes the agent to write files, perform refactoring, and execute local build/lint tools. * Sanitization: There is no evidence of sanitization or filtering to prevent the agent from following instructions embedded within the processed code data.

code-craft