The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes untrusted data in the form of CLR activation logs (.CLRLoad*.log) provided by the user to diagnose activation failures.
Ingestion points: Activation log files identified in the workflow (Workflow Step 2).
Boundary markers: The skill does not explicitly use delimiters to separate log content from instructions, though the logs are system-generated and structured.
Capability inventory: Analysis is limited to searching logs via grep and reading local system state (registry keys and file presence). No file-write, network exfiltration, or dynamic code execution capabilities are triggered by the log content.
Sanitization: No specific sanitization or escaping of log content is implemented during the diagnostic process.
Risk Assessment: The risk is negligible as the skill only provides informational diagnosis to the user without executing autonomous actions based on the content of the logs.
[SAFE]: The skill follows secure patterns by directing the agent to use bundled reference materials from its own installation directory rather than relying on potentially untrusted workspace files. All referenced tools (e.g., fondue.exe) and registry paths (e.g., HKLM\SOFTWARE\Microsoft\.NETFramework) are standard components of the Windows operating system and .NET Framework environment. No evidence of obfuscation, credential harvesting, or unauthorized remote access was found.

clr-activation-debugging