The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill interpolates user-supplied input via the $ARGUMENTS variable directly into its operational instructions to define the audit scope.
Ingestion points: The $ARGUMENTS variable is used in the SCOPE section of SKILL.md.
Boundary markers: The instructions do not employ delimiters or specific 'ignore' directives to separate the user-provided scope from the agent's core instructions.
Capability inventory: The agent performs recursive codebase scanning, reads project data files (TMX, JSON, CSV), and has the capability to write telemetry logs to the user's home directory.
Sanitization: No evidence of validation, escaping, or sanitization of the $ARGUMENTS value was found.
[PERSISTENCE]: The 'Self-Evolution Telemetry' section instructs the agent to maintain state across sessions by writing to a hidden directory outside the project root.
Evidence: The skill defines a process to search for the path ~/.claude/projects/ and append execution metadata to skill-telemetry.md. While used for tracking skill performance ('evolution'), this constitutes an automated mechanism for persisting data in a hidden system directory.

level-design