The Agent Skills Directory

[PROMPT_INJECTION]: The skill implements an autonomous mode (--interactive false) that explicitly instructs the agent to skip all user approval gates and decision points. This behavior extends to critical operations such as resolving logic ambiguities and deploying enhanced scripts to a Salesforce organization, effectively removing human oversight from the deployment pipeline.
[INDIRECT_PROMPT_INJECTION]: The skill parses Einstein Bot XML metadata and Apex source code to generate structured prompts for the /agentforce-generate tool. This process lacks explicit safeguards to prevent instructions within the ingested data from influencing the agent's behavior.
Ingestion points: Einstein Bot metadata files (.bot-meta.xml, .botVersion-meta.xml), Apex classes, and Flow artifacts retrieved from the Salesforce org or a local directory as described in references/extraction-blueprint.md.
Boundary markers: There are no documented boundary markers or instructions to ignore embedded commands when the extracted content is passed to the /agentforce-generate agent in SKILL.md (Step 4).
Capability inventory: The skill is capable of executing shell commands via the sf CLI, writing local files, and deploying code to Salesforce environments (references/post-conversion-enhancements-reference.md).
Sanitization: No specific sanitization, escaping, or filtering logic is defined for handling the content extracted from the source metadata before prompt interpolation.
[COMMAND_EXECUTION]: The skill relies on the Salesforce CLI (sf) to manage project files, query org data, and retrieve or deploy metadata. While these are standard developer actions, the commands are dynamically constructed using bot names and version strings derived from external metadata.

agentforce-bot-upgrade