The Agent Skills Directory

[PERSISTENCE]: The skill provides examples and instructions for configuring crontab jobs to automate AI workflows. While presented as a core feature for scheduled tasks, these are standard persistence mechanisms that allow code to execute automatically across sessions.
[INDIRECT_PROMPT_INJECTION]: The data_processing.sh template reads content from local files and injects it directly into a prompt using $(cat $file). This pattern is vulnerable to indirect prompt injection as it lacks boundary markers or sanitization to prevent data from being interpreted as instructions by the downstream AI model.
Ingestion points: data_processing.sh reads files from the ./data/raw directory.
Boundary markers: None identified. Data is appended directly to the end of the prompt string.
Capability inventory: The skill utilizes command execution via belt, file system operations (mkdir, cat), and network requests (curl).
Sanitization: No sanitization or escaping is applied to the file content before it is processed.
[DATA_EXFILTRATION]: The monitored_workflow.sh template includes a curl command that posts execution data to an external URL (https://your-webhook.com/alert). This demonstrates how the skill can be configured to transmit data from the local environment to external servers.
[COMMAND_EXECUTION]: The provided Python script (automation.py) uses subprocess.run to call the belt CLI, representing a pattern of dynamic command execution within the workflow.
[EXTERNAL_DOWNLOADS]: The skill documentation encourages the installation of external components via npx skills add and references installation scripts hosted on GitHub (raw.githubusercontent.com/inference-sh/skills).

ai-automation-workflows