The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to perform environment discovery using shell commands such as test, command -v, sf, and python3. These commands are used to determine which "execution mode" the agent is running in and which local tools are available for processing Salesforce data.
[DYNAMIC_EXECUTION]: The skill documentation describes execution modes that rely on running local code (Python, shell, jq) for post-processing and report generation. It specifically instructs the agent to locate validation scripts by searching the user's home directory (find ~/ -name "<script_name>.py") if environment variables like CLAUDE_PLUGIN_ROOT are not set. This represents the execution of scripts from dynamically discovered/computed paths.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external web sources (research findings, LinkedIn profiles) to enrich Salesforce records, which creates a vulnerability surface for malicious instructions embedded in third-party content.
Ingestion points: External websites, company pages, and LinkedIn profiles as specified in the research workflow in SKILL.md.
Boundary markers: Absent; no explicit instructions are provided to the agent to treat external content as untrusted or to use delimiters.
Capability inventory: The skill utilizes soql_query for reading and sobject_dml (via Salesforce MCP tools) for writing data to the CRM.
Sanitization: The skill recommends mapping data to the "org's Industry picklist values" and performing "field-length checks," but lacks comprehensive sanitization for natural language data before it enters the agent's context.

sf-leads