The Agent Skills Directory

[COMMAND_EXECUTION]: The skill automatically executes brew install to install system packages like yt-dlp if they are not detected. This modifies the host system environment without explicit user confirmation or a separate setup phase.
[COMMAND_EXECUTION]: The skill uses shell scripts to process user-provided URLs directly in commands such as curl, yt-dlp, and reader. There is no evidence of shell escaping or sanitization for the $URL variable, which could lead to command injection if a malicious URL (e.g., containing backticks or semicolons) is provided.
[EXTERNAL_DOWNLOADS]: The skill downloads content from arbitrary user-provided URLs using curl, yt-dlp, and other tools. While this is the primary purpose of the skill, it creates a vector for interacting with malicious servers.
[PROMPT_INJECTION]: The skill is highly vulnerable to Indirect Prompt Injection. It ingests untrusted data from external sources (YouTube transcripts, web articles, PDFs) and passes this content to the ship-learn-next skill to create an action plan. If the source content contains malicious instructions designed to manipulate the agent, the agent may follow them during the planning phase.
Ingestion points: Content extracted from URLs and saved to .txt files in Step 2.
Boundary markers: None. The content is read directly from the file into the planning context.
Capability inventory: The skill has Bash, Read, and Write permissions, allowing it to execute system commands and modify files based on the generated plan.
Sanitization: None. The script performs basic HTML/text cleaning for formatting but does not filter for instructional content.

tapestry