The Agent Skills Directory

[COMMAND_EXECUTION]: The module scripts/auth.py executes the Salesforce CLI (sf) via subprocess.run to obtain session tokens and instance URLs for API authentication. This behavior is expected for local developer tools that leverage existing authenticated CLI sessions.
[DYNAMIC_EXECUTION]: The permission_detector.py and tooling_api.py scripts build SOQL queries using string concatenation. While common in Salesforce tooling, this pattern creates a theoretical surface for SOQL injection if input identifiers are maliciously crafted.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and displays metadata from external Salesforce orgs, which could potentially house adversarial instructions. 1. Ingestion points: External data enters through SOQL queries in permission_detector.py, user_analyzer.py, and hierarchy_viewer.py. 2. Boundary markers: The output uses basic markdown formatting and Mermaid delimiters, but lacks explicit guardrails to prevent an LLM from interpreting metadata content as commands. 3. Capability inventory: The skill can execute shell commands, perform network requests to Salesforce, and write analysis results to local files. 4. Sanitization: The skill performs syntax-specific escaping for Mermaid diagrams in scripts/renderers/mermaid.py.

sf-permissions