The Agent Skills Directory

[SAFE]: The skill implements a benchmarking workflow for Salesforce development. It operates on local files within the defined project structure and does not perform network operations or access sensitive system credentials.
[COMMAND_EXECUTION]: The skill executes a local shell script (evals/checks/static-checks.sh) to perform static analysis on Salesforce code. This is a functional requirement of the skill and uses a relative path to a script included in the skill package.
[DATA_EXFILTRATION]: No data exfiltration patterns or access to sensitive configuration files (e.g., .env, .ssh, .aws) were identified. The skill's file operations are limited to reading benchmarking configurations and writing results to a local results directory.
[INDIRECT_PROMPT_INJECTION]: The skill ingests external data from evals/benchmarks/tasks.json and user-provided code samples (Ingestion Points). No explicit boundary markers are defined in the prompt text. The skill utilizes Write, Edit, and Bash capabilities across its scripts (Capability Inventory). No specific sanitization or escaping of external content is defined. While this presents an attack surface common to tools processing external data, the risk is inherent to the primary benchmarking purpose and is managed by the agent's internal logic.

sf-eval