The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill recommends executing a command that downloads and runs code directly from an untrusted third-party repository (github.com/laiso/site2skill) using uvx. This allows for the execution of arbitrary remote code on the host system.
[EXTERNAL_DOWNLOADS]: Fetches the official Claude Code CHANGELOG.md from the anthropics/claude-code repository using the GitHub CLI to assist with version tracking and user queries.
[COMMAND_EXECUTION]: Employs several local shell commands to facilitate configuration tasks, including claude --version for environment detection, gh api for remote data retrieval, and base64 for processing responses.
[PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection as it processes externally-sourced changelog data.
Ingestion points: The skill downloads the CHANGELOG.md file from the public anthropics/claude-code GitHub repository (SKILL.md, Section 2.2).
Boundary markers: No specific delimiters or instructions are used to ensure the agent ignores potentially adversarial instructions embedded within the fetched changelog content.
Capability inventory: The agent has access to the Bash tool, enabling it to execute system commands and modify configuration files (SKILL.md).
Sanitization: External content is decoded and processed without validation or sanitization before being provided to the agent context.

claude-config-optimizer