The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the aws CLI, jq, find, and sed utilities within the publish.sh script to manage AWS resources and validate file structures. Evidence: scripts/publish.sh lines 145-156.
[DATA_EXFILTRATION]: The skill is designed to upload the contents of a local directory to a remote S3 bucket. To mitigate the risk of exfiltrating sensitive system files, the script explicitly prohibits hidden files, dot-paths, and symlinks. Evidence: scripts/publish.sh lines 173-193.
[EXTERNAL_DOWNLOADS]: The publish.sh script performs a curl request to the newly published URL to verify availability. Evidence: scripts/publish.sh line 485.
[PROMPT_INJECTION]: The skill has an indirect prompt injection surface as it ingests local files for upload without adding specific boundary markers to the content. Evidence Chain: (1) Ingestion Point: scripts/publish.sh reads the local directory tree; (2) Boundary Markers: Absent; (3) Capability Inventory: aws s3 sync, curl, aws s3api; (4) Sanitization: The script enforces a root index.html and rejects all hidden files and symlinks. Evidence: scripts/publish.sh ingestion at line 472; sanitization at lines 173-193.

paperclip-page