The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a suite of Node.js and Bash scripts (e.g., scripts/deploy_and_run_recipe.js, scripts/check_connector_error.sh) that execute system commands to interface with the Salesforce CLI (sf) and perform network operations via curl to interact with the Salesforce REST API.
[DYNAMIC_EXECUTION]: Several scripts, such as scripts/create_app.js and scripts/upload_dataset.js, dynamically construct and execute shell commands at runtime using execSync. While these scripts implement parameter escaping for curl arguments, they represent a dynamic execution surface within the skill's operational flow.
[INDIRECT_PROMPT_INJECTION]: The skill acts as an ingestion point for Salesforce metadata, specifically SAQL queries from dashboards and Wave Recipe JSON definitions. An adversary with access to the Salesforce org could embed malicious instructions within these assets to influence the agent's behavior when it audits or deploys them.
Ingestion points: scripts/find_unused_dataset_fields.js reads dashboard SAQL queries; scripts/deploy_and_run_recipe.js reads Wave Recipe definitions.
Capability inventory: File system read/write, network access to Salesforce APIs (via curl), and shell command execution.
Boundary markers: No explicit delimiters or safety instructions are used when interpolating these external data sources into the agent's context.
Sanitization: The scripts perform HTML entity decoding for functional reasons, but do not provide security-focused sanitization or filtering of the content.

crm-analytics-wave-generate