The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill dynamically creates and modifies Salesforce metadata files (XML and YAML) at runtime. It retrieves templates from the assets/ directory and interpolates user-supplied values, such as agent labels, queue IDs, and routing labels, before deploying them to the target organization using the sf project deploy command. This behavior is found in SKILL.md, references/channel-branch-email.md, and references/routing-flow.md.
[INDIRECT_PROMPT_INJECTION]: The skill provides a surface for indirect injection by ingesting untrusted data through user prompts and inserting it into configuration files.
Ingestion points: User-provided strings for LEGAL_DISCLAIMER_TEXT, SIGNATURE_TEXT, and LABEL are collected in references/channel-branch-email.md.
Boundary markers: The instructions do not specify any delimiters or warnings to ignore embedded instructions within these text fields.
Capability inventory: The skill has the capability to deploy these modified files to a Salesforce org via sf project deploy start and sf api request rest (documented in references/channel-branch-email.md).
Sanitization: There is no evidence of sanitization or escaping for XML/YAML special characters in the provided Python validation scripts or the main workflow instructions.

service-agentforce-channel-configure