The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The scripts/view_access_logs.sh script downloads log files from a DigitalOcean Spaces bucket and uses grep to filter their content. Access logs contain data from external requests, such as User-Agent strings or request URIs, which are controlled by external users. If an attacker can influence these logs, they could potentially inject malicious instructions that the AI agent might follow when it processes the log output.
Ingestion points: The scripts/view_access_logs.sh script ingests content from external Spaces buckets using aws s3 sync.
Boundary markers: No specific delimiters or "ignore embedded instructions" warnings are used in the scripts to isolate log content from the agent's instructions.
Capability inventory: The skill has broad capabilities, including creating and deleting access keys (doctl spaces keys), managing buckets, and performing object operations. These capabilities could be targeted if an agent obeys instructions found within the logs.
Sanitization: The script performs no sanitization or escaping of the log data before it is presented to the user or agent output.
[EXTERNAL_DOWNLOADS]: The documentation in reference/aws-cli-operations.md provides instructions for downloading and installing the AWS CLI from Amazon's official domain (awscli.amazonaws.com). This is documented neutrally as it targets a well-known and trusted technology provider.

spaces