The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's 'audit' functionality involves reading and analyzing existing project files to evaluate their compliance with SGR patterns. This creates an inherent surface for indirect prompt injection if the audited files contain adversarial instructions. However, this behavior is central to the skill's purpose as a code analysis tool.
Ingestion points: $ARGUMENTS variable and project source files (e.g., Python Pydantic models, Zod schemas) during audit mode.
Boundary markers: The instructions do not define specific delimiters or guardrails to prevent the agent from obeying instructions embedded within the code it audits.
Capability inventory: The skill utilizes Read, Write, Edit, and Bash tools to analyze and potentially modify the project structure.
Sanitization: No explicit sanitization or filtering of audited file content is implemented.
[SAFE]: The Python simulation provided in references/sgr-demo.py and references/sgr-full-guide.md is a benign demonstration of agentic tool dispatch. It uses an in-memory database and performs no external network or sensitive file system operations beyond the intended interaction with the OpenAI API for structured outputs.

solo-sgr