The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands exclusively through the official Salesforce CLI (sf) and node. All command arguments are carefully managed; large payloads (like Agent Scripts) are processed by specialized Node.js scripts and passed to the CLI via file references (--body @<file>) rather than inline strings, which mitigates shell injection risks.
[CREDENTIALS_SAFE]: The skill explicitly forbids the extraction of access tokens. It relies on the Salesforce CLI's internal session management (sf api request rest), ensuring that sensitive bearer tokens are never exposed in the shell environment or logs.
[DATA_EXFILTRATION]: All network operations are directed at official Salesforce API endpoints (/services/data/v67.0/...) using the authenticated CLI session. No data is sent to unauthorized or external third-party domains.
[OBFUSCATION]: The helper scripts include HTML entity decoding logic. This is a functional requirement for correctly processing Salesforce template data (which is returned double-encoded by the API) and is not used to hide malicious intent or commands.
[PRIVILEGE_ESCALATION]: No commands requiring elevated system privileges (such as sudo or chmod 777) are present. The skill operates within the context of the authenticated Salesforce user's permissions.
[DYNAMIC_EXECUTION]: The skill uses local Node.js scripts bundled within its directory to perform deterministic logic. These scripts use standard built-in modules (fs, path) and do not perform unsafe execution of untrusted code or deserialization of untrusted data.
[INDIRECT_PROMPT_INJECTION]: While the skill ingests data from Salesforce APIs (like agent templates), it uses strict regex-based extraction and deterministic logic in its classifiers. It also includes a mandatory AskUserQuestion gate to ensure the user reviews and approves the configuration before any write operation occurs.

service-itsm-agentic-setup-fulfiller-agent-configure