filesystem-context

Pass

Audited by Gen Agent Trust Hub on Mar 7, 2026

Risk Level: SAFENO_CODE
Full Analysis
  • [SAFE]: The skill consists entirely of markdown documentation. It provides conceptual patterns and pseudo-code examples for context engineering but does not ship with any functional code or executable scripts.
  • [INDIRECT_PROMPT_INJECTION]: The guide describes patterns for ingesting external data (such as web search results or tool outputs) into the filesystem for the agent to read later. This creates a potential surface for indirect prompt injection if the ingested data contains malicious instructions.
  • Ingestion points: Pattern 1 (Tool Output Offloading) and Pattern 5 (Terminal/Log Persistence) describe saving external data to files.
  • Boundary markers: Examples show returning summaries and file references to the context window, but do not provide specific instruction delimiters for the files themselves.
  • Capability inventory: The skill assumes the agent has access to write_file, read_file, grep, ls, and glob tools.
  • Sanitization: Not provided in the documentation; however, the guide includes a warning to 'guard self-modification patterns with validation'.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 7, 2026, 05:35 PM