The Agent Skills Directory

[DATA_EXFILTRATION]: The read_file function in scripts/helper.py accepts a file path string and reads it without performing any validation or sanitization. This implementation allows for arbitrary file access, enabling the retrieval of sensitive system files, configuration files, or credentials if requested.
[COMMAND_EXECUTION]: The YAML frontmatter in SKILL.md requests the wildcard permission *, as well as bash and network permissions. These privileges are over-scoped for a utility described as a simple file reader and grant the skill broad authority to execute commands and interact with the network.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface by ingesting data from external files into the agent context without sanitization. If the agent reads a file containing malicious instructions, it may attempt to execute them using its elevated permissions.
Ingestion points: The read_file function in scripts/helper.py reads local file content.
Boundary markers: Absent. The skill does not wrap file content in delimiters or provide warnings to the agent.
Capability inventory: The skill possesses bash, network, write, and wildcard * permissions.
Sanitization: None. File content is passed directly to the agent as a raw string.

over-privileged-helper