The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands using python3 to run internal scripts like eva_link_check.py and eva_asset_validate.py. These commands interpolate absolute paths (EVA_SHARED_ROOT, PROJECT_ROOT) and user-defined Link IDs into the shell strings, which could lead to command injection if paths are not properly sanitized or if the environment is compromised.
[PROMPT_INJECTION]: The skill processes 'local modules' which include module.md files containing natural language instructions (SOPs and methodologies). This represents an Indirect Prompt Injection surface where malicious instructions could be embedded in the processed data. The skill implements a fingerprinting mechanism (approved_sha256) and 'strict' validation to detect 'carried instructions' (夹带指令) to mitigate this risk.
[DATA_EXFILTRATION]: While the default configuration for local links disables network access (network: false), the framework's ability to read files from the local filesystem (.eva/links.json, local-modules/) and execute scripts could be leveraged for data exposure if the security boundaries are bypassed.
[DYNAMIC_CONTEXT_INJECTION]: The skill explicitly defines a mechanism to resolve and use absolute paths for script execution at runtime to avoid directory traversal issues, which is a defensive measure but highlights the dynamic nature of the command execution environment.

eva-link