The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to construct and execute shell commands using the crw CLI tool. There is a potential risk of command injection if the agent interpolates unvalidated user search queries directly into shell strings without proper escaping.
[REMOTE_CODE_EXECUTION]: The skill relies on generating and executing dynamic Python logic using heredocs (python3 << 'PYEOF') and command-line execution (python3 -c). While this is intended for data processing and token optimization, it involves the runtime generation of executable code.
[PROMPT_INJECTION]: The skill possesses a significant attack surface for indirect prompt injection as it is designed to ingest and process data from arbitrary external websites.
Ingestion points: Web data is retrieved via crw search and crw scrape (SKILL.md).
Boundary markers: While the skill promotes "context isolation" by processing data in a Python subprocess, it lacks specific sanitization logic to prevent malicious instructions embedded in scraped Markdown or HTML from being printed to the context window.
Capability inventory: The skill utilizes the Bash tool to execute python3 and crw, and stores intermediate data in the /tmp/ directory.
Sanitization: No explicit sanitization or filtering of adversarial text (e.g., hidden instructions) is performed on the scraped content before it is returned to the agent context.

crw-dynamic-search