The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/search_token.sh script is vulnerable to Python code injection. The --token argument provided by the user is interpolated directly into a Python heredoc (token = "${TOKEN}".upper()) inside the bash script. An attacker can supply a crafted token string containing quotes and Python commands (e.g., BTC"; import os; os.system("id"); #) to execute arbitrary code with the privileges of the agent.
[DYNAMIC_EXECUTION]: Multiple scripts in the skill (scripts/search_token.sh, scripts/test_all.sh, and scripts/test_connector.sh) use bash-to-python interpolation to perform data processing. This architectural pattern of generating executable code from shell variables at runtime is highly insecure and creates multiple vectors for exploitation when handling unsanitized inputs.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits a significant attack surface for indirect prompt injection as it ingests data from a remote API and local configuration files, then processes and displays this data using unsafe execution methods.
Ingestion points: Data is ingested from the Hummingbot API via curl in scripts/test_all.sh and scripts/test_connector.sh, and subsequently read from data/trading_rules.json in scripts/search_token.sh.
Boundary markers: None. The skill assumes all data returned by the configured API and provided by the user is well-formed and safe.
Capability inventory: The skill has the capability to execute shell commands (bash), make network requests (curl), and execute arbitrary Python code (python3).
Sanitization: There is no sanitization of the TOKEN variable or the API responses before they are interpolated into executable script blocks.

connectors-available