The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: The skill processes AWS credentials for S3 uploads. These are sourced exclusively from environment variables (EVIDENCE_AWS_ACCESS_KEY_ID, EVIDENCE_AWS_SECRET_ACCESS_KEY), which is a standard and secure management practice. No credentials are hardcoded.
[COMMAND_EXECUTION]: The skill executes shell commands (curl, git, sha256sum) which are necessary for its documented functionality. It implements robust sanitization of inputs such as repository names and path prefixes to prevent injection or unexpected behavior.
[INDIRECT_PROMPT_INJECTION]: The skill ingests file paths and metadata (alt text, prefixes) that could theoretically contain malicious content. However, the script performs regex-based sanitization and avoids unsafe operations like eval, effectively mitigating the risk of command injection via these inputs.
[SAFE]: The skill utilizes a content-addressed storage approach for visual assets. It includes a clear warning about the public nature of the uploaded URLs, advising against the upload of sensitive data or PII.

evidence-upload