The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructions include a command to download and execute a shell script from a remote URL: curl -fsSL https://cli.inference.sh | sh. This 'curl piped to shell' pattern is a high-risk activity as it executes unverified code from an untrusted source with the user's shell privileges.- [COMMAND_EXECUTION]: The skill uses the infsh CLI tool to run various remote-hosted applications. While the skill configuration attempts to restrict usage to this tool, the underlying capability allows the agent to trigger complex external operations.- [DYNAMIC_EXECUTION]: The skill utilizes an infsh/python-executor tool to run Python code blocks generated at runtime (e.g., for creating positioning maps). This allows for the execution of arbitrary logic, which could be repurposed for malicious use if the prompt generation is compromised.- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest data from third-party websites and search results to perform its analysis, creating a significant attack surface.
Ingestion points: Website content accessed via agent-browser and tavily/extract, and search results from tavily/search-assistant.
Boundary markers: None identified. The skill does not instruct the agent to use delimiters or ignore instructions embedded within the retrieved competitive data.
Capability inventory: The agent has access to shell commands (via infsh), Python execution, and network request capabilities.
Sanitization: No sanitization or validation of the data retrieved from external URLs is specified in the skill's instructions.- [EXTERNAL_DOWNLOADS]: The skill fetches binaries and CLI tools from inference.sh domains. As these are not recognized as trusted or well-known services, this presents a potential supply-chain risk.

competitor-teardown