The Agent Skills Directory

[SAFE]: No malicious patterns, obfuscation, or unauthorized data access were detected. The skill instructions prioritize system stability and data integrity by enforcing read-only assessments and requiring explicit authorization for any modifications. All external documentation links refer to official Microsoft domains.
[INDIRECT_PROMPT_INJECTION]: The skill has a potential attack surface through the ingestion of system telemetry and logs, which could theoretically be manipulated by local malicious actors. This risk is managed by the following controls:
Ingestion points: System diagnostic outputs including Task Manager process lists, Event Viewer logs, and Defender detection history.
Boundary markers: Comprehensive safety boundaries and prohibited operations are explicitly defined in the documentation (references/safety-boundaries.md) to prevent the agent from following unsafe instructions embedded in data.
Capability inventory: The skill uses authorized Windows tools (DISM, SFC, CHKDSK, defrag, powercfg) and managed file cleanup via system-supported mechanisms.
Sanitization: The workflow mandates a 'preview and confirm' step for all actions, ensuring a human-in-the-loop reviews every proposed mutation before execution.

windows-cleanup