The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input through post text and image file paths, which could be leveraged to perform unauthorized file reads if the agent is targeted by a malicious prompt.
Ingestion points: The functions publishPostWithImage and bulkSchedule in SKILL.md accept text and imagePath parameters from external inputs.
Boundary markers: The instructions lack explicit delimiters or instructions for the agent to ignore potentially malicious embedded commands within the processed data.
Capability inventory: The skill utilizes fs.readFileSync for local file access and the fetch API for network transmission to the Hootsuite platform.
Sanitization: No path validation or sanitization is implemented to restrict the imagePath to specific directories.
[DATA_EXFILTRATION]: The skill facilitates the transmission of local file content to the Hootsuite REST API (platform.hootsuite.com). While the destination is a well-known service and the workflow is legitimate, the capability to read and transmit local files based on dynamic input represents a data exposure surface.

hootsuite-core-workflow-a