The Agent Skills Directory

[COMMAND_EXECUTION]: The helper script scripts/research.sh is vulnerable to command injection. The KEYWORD and SEARCH_TERM variables, which are provided by the user, are interpolated directly into a Python command string (python3 -c) using single quotes. A malicious keyword containing a single quote could break out of the intended Python string context and execute arbitrary Python or shell commands on the host system. Evidence: ENCODED_KW=$(python3 -c "import urllib.parse; print(urllib.parse.quote('$KEYWORD'))") in scripts/research.sh.
[DYNAMIC_EXECUTION]: The skill uses shell scripts to dynamically generate and execute Python code snippets for utility tasks like URL encoding and JSON parsing. The lack of input sanitization during this dynamic code assembly constitutes a security risk.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources that can be influenced by third parties. 1. Ingestion points: Amazon Autocomplete API responses and web search results for competitors and price patterns. 2. Boundary markers: The skill instructions do not define delimiters or specific safety warnings to the agent to ignore instructions embedded within the retrieved data. 3. Capability inventory: The skill has access to shell script execution, web search, and web fetch tools. 4. Sanitization: The skill does not validate or sanitize the data retrieved from Amazon or search engines before presenting it to the agent.

amazon-keyword-research