The Agent Skills Directory

[PROMPT_INJECTION]: Indirect Prompt Injection Surface
The skill's workflow involves gathering business requirements (Phase 1) that are subsequently used to generate OmniScript JSON and populate parameters for deployment scripts. This creates a surface for indirect prompt injection where malicious instructions in the requirements could influence the agent's behavior.
Ingestion points: Requirement inputs such as 'Type', 'SubType', and 'Purpose' defined in Phase 1 (SKILL.md).
Boundary markers: The skill lacks explicit instructions or delimiters to isolate user-provided requirements from the agent's internal logic.
Capability inventory: The skill executes shell scripts (scripts/deploy-omniscript.sh), uses the Salesforce CLI for data queries and record deletion, and generates system metadata.
Sanitization: No input validation or escaping mechanisms are defined for the requirements before they are used in script execution or metadata generation.
[COMMAND_EXECUTION]: Shell Parameter Interpolation in Helper Scripts
Several shell scripts, including scripts/check-duplicate-omniscript.sh and scripts/deploy-omniscript.sh, utilize direct string interpolation of variables (e.g., ${TYPE}, ${SUBTYPE}, ${NAME}) within shell commands.
Specifically, scripts/deploy-omniscript.sh contains a command to delete records: sf data delete record -s OmniProcess -w \"Name='${NAME}'\" -o ${ORG}. If the variable ${NAME} is derived from untrusted user input without sanitization, it could lead to SOQL injection or unintended record deletion.

building-omnistudio-omniscript