The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs users to install the fabric CLI by piping a remote shell script from a GitHub repository directly into the bash interpreter (curl -fsSL ... | bash), allowing for the execution of unverified code with the user's current system privileges.
[EXTERNAL_DOWNLOADS]: The fabric -u command is used to download and update a library of AI patterns from a remote repository at runtime without integrity checks, introducing unvetted external content into the local environment.
[COMMAND_EXECUTION]: The skill relies extensively on shell command execution and piping to process data, including instructions to read content from the system clipboard (pbpaste) and system logs.
[CREDENTIALS_UNSAFE]: The setup process involves configuring sensitive API keys for multiple AI providers (fabric --setup), which the skill stores in an unencrypted local configuration file (~/.config/fabric/.env).
[PROMPT_INJECTION]: The skill processes untrusted external data from web URLs and YouTube transcripts by interpolating it into AI prompts. Ingestion points: 'SKILL.md' (Steps 2, 4, and 7) using curl, fabric -y, and Unix pipes. Boundary markers: Absent; no delimiters or warnings are used to segregate external content. Capability inventory: 'SKILL.md' (allowed-tools) includes Bash, Write, and Read. Sanitization: Absent; no evidence of input validation or escaping before data is processed by the AI patterns.

fabric