filesystem-context

Pass

Audited by Gen Agent Trust Hub on Mar 10, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill introduces a surface for indirect prompt injection via 'Self-Modification' and 'Dynamic Skill Loading' patterns.
  • Ingestion points: Untrusted data from sources like web searches or database queries are written to the filesystem in directories like scratch/ and memory/ as described in SKILL.md.
  • Boundary markers: No specific delimiters or boundary instructions are provided to isolate the ingested external data from the agent's core instructions or static context.
  • Capability inventory: The system leverages file-system tools including read_file, write_file, grep, and glob to manage and retrieve this persisted context.
  • Sanitization: While the documentation notes that 'self-modification requires careful guardrails', it provides no concrete sanitization or validation logic to prevent malicious content from being interpreted as instructions.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 10, 2026, 01:39 AM