The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted project source code as data to identify issues for removal. This creates a potential surface for indirect prompt injection, where content within project files could attempt to influence the agent's behavior. However, the risk is minimal as the skill's capabilities are strictly limited to removing specific, predefined code patterns.
Ingestion points: Project source files accessed via the sniff MCP tool or legacy sniff_scan tool.
Boundary markers: The skill does not explicitly define delimiters for findings, but instructions constrain the agent to a specific list of fixable issues.
Capability inventory: File modification through the removal of specific code lines.
Sanitization: The skill implements a whitelist approach by only authorizing the removal of specific patterns like debug-debugger and debug-console-log while requiring manual intervention for other findings.

sniff-fix