The Agent Skills Directory

[COMMAND_EXECUTION]: The skill interpolates the $ARGUMENTS variable directly into several shell commands (e.g., just validate-modules $ARGUMENTS, just analyze-ranges $ARGUMENTS). This creates a command injection surface if the module name provided by a user contains shell metacharacters or malicious payloads.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface (Category 8) where instructions from processed data could potentially influence agent behavior.
Ingestion points: Untrusted data enters the context via user-provided $ARGUMENTS and the output from sub-agents (e.g., {VERIFIER_OUTPUT}, {JUDGE_OUTPUT}).
Boundary markers: The skill does not use delimiters or explicit instructions to treat the interpolated data as untrusted when forming prompts for the Judge or Dev agents.
Capability inventory: The skill has the capability to execute arbitrary shell commands via the just runner, modify local files (DSP source and JSON configs), and spawn additional agents with new prompts.
Sanitization: There is no visible evidence of input validation or output sanitization before the data is used to construct shell commands or sub-agent instructions.

module-dev