The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits a vulnerability to indirect prompt injection by processing untrusted data from project.config.json to guide code generation and shell command execution.
Ingestion points: The skill reads transport settings from project.config.json at runtime.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to ignore embedded instructions within the configuration file.
Capability inventory: The skill utilizes Write, Edit, and Bash tools to scaffold the communication layer, which could be exploited to write malicious files or execute arbitrary commands if the configuration is poisoned.
Sanitization: No evidence of validation or sanitization of the configuration data is present in the skill instructions.
[DATA_EXFILTRATION]: The skill documentation exposes sensitive environment-specific information by referencing absolute local file system paths.
Evidence: The file explicitly references D:\FSO\Computer Networks as a source directory, disclosing the host's directory structure and potential user environment details.

spaghetti-relay-deploy