The Agent Skills Directory

COMMAND_EXECUTION (LOW): The skill utilizes shell scripts (scripts/search.sh, scripts/content.sh, scripts/code.sh) and requires the installation of curl and jq. While inputs are passed through jq to prevent shell injection, the execution of shell scripts increases the overall attack surface.
DATA_EXFILTRATION (MEDIUM): The scripts search.sh and content.sh access the local file system at ~/.clawdbot/credentials/exa/config.json to retrieve an API key. This constitutes access to sensitive credential data, which is subsequently transmitted to the external domain api.exa.ai for authentication.
PROMPT_INJECTION (MEDIUM): The skill is susceptible to Indirect Prompt Injection (Category 8) because it fetches and processes untrusted data from the internet.
Ingestion points: External search results and website text are retrieved via the Exa AI API and presented to the agent in scripts/search.sh, scripts/content.sh, and scripts/code.sh.
Boundary markers: None; the scripts do not implement delimiters or instructions to prevent the agent from obeying commands embedded within the search results.
Capability inventory: The skill possesses network communication capabilities via curl to reach api.exa.ai.
Sanitization: While the scripts use jq to ensure valid JSON payloads are sent to the API, there is no evidence of sanitization or filtering of the content returned from the external API before it is processed by the agent.

exa-plus