macos-storage-cleanup

Pass

Audited by Gen Agent Trust Hub on Aug 24, 2026

Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes file system metadata and tool outputs which are technically untrusted inputs. However, it mitigates potential injection or coercion risks by enforcing a strict "preview-then-confirm" workflow where every action is itemized and requires a separate user interaction turn.
  • Ingestion points: File paths, directory sizes, and CLI tool metadata (SKILL.md, cleanup-catalog.md).
  • Boundary markers: Mandatory itemized preview table with stable row identifiers and a requirement for fresh, explicit consent per row.
  • Capability inventory: Targeted file deletion, cache pruning, and service management via established system binaries.
  • Sanitization: Resolution of targets through owning tools, remeasuring before action, and explicit refusal to follow unexpected symlinks.
  • [COMMAND_EXECUTION]: The skill uses local command-line tools and system binaries (such as tmutil, docker, and npm) to perform audits and cleanup. These operations are governed by a static catalog of commands and flags, ensuring that the agent does not invent dangerous parameters or perform blind recursive deletions.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 24, 2026, 01:42 AM
Security Audit — agent-trust-hub — macos-storage-cleanup