The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the opentaint scan command-line utility. The command incorporates variables such as <test-model> and <config-file> which are derived from user input or project context, presenting a risk of command injection if these inputs are not properly sanitized.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface (Category 8). It ingests untrusted data describing library methods and interpolates this into a generated YAML configuration file.
Ingestion points: The skill reads method propagation data from a tracking file and user-provided descriptions.
Boundary markers: No explicit delimiters or boundary markers are used to isolate untrusted input within the generated configuration.
Capability inventory: The skill can write files to the local file system (<config-file>) and execute the opentaint scan shell command.
Sanitization: There is no evidence of input validation or escaping for the parameters passed to the shell command or the configuration generator.
[DATA_EXFILTRATION]: The skill reads and writes data to paths within the .opentaint/ directory. While these operations are local to the project workspace, the skill is designed to process and propagate information about code behavior (taint tracking), which involves reading project metadata.

create-pass-through-approximation