The Agent Skills Directory

[COMMAND_EXECUTION]: The skill interacts with the Salesforce environment using standard MCP tools such as sobject_describe, soql_query, and tooling_api_query. These capabilities are used exclusively for legitimate metadata discovery required to build accurate data model diagrams (ERDs).
[INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it ingests untrusted data from the Salesforce org (e.g., object descriptions, field labels, and metadata values). While this data is interpolated into prompts to generate diagrams, the risk is mitigated by the skill's specific focus on generating Mermaid and ASCII text output, and the instructions to wait for user input rather than defaulting to guesses.
[EXTERNAL_DOWNLOADS]: The skill handles large responses (artifacts) from the MCP server by downloading them from a signed artifactUrl when the environment supports local file writing. This is a standard and expected mechanism within the Model Context Protocol (MCP) for handling payloads that exceed context window limits.
[SAFE]: The skill explicitly defines the removal of legacy developer scripts and relies entirely on natural language instructions and pre-defined markdown templates. No hardcoded credentials or malicious persistence mechanisms were detected. The use of placeholders like CONSUMER_KEY and CONSUMER_SECRET in OAuth templates is clearly documented for user documentation purposes and does not represent a credential exposure risk.

sf-diagram