content-hash-cache-pattern
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill provides a pattern for reading and processing external file content (e.g., PDF parsing, text extraction). This creates a surface for indirect prompt injection where malicious instructions inside a processed file could be executed by the agent if the results are directly interpolated into the prompt.
- Ingestion points: The compute_file_hash and extract_with_cache functions in SKILL.md read file content from the local filesystem.
- Boundary markers: The provided code snippets do not include delimiters or specific warnings to ignore instructions within the file content.
- Capability inventory: The pattern involves file system read access for hashing and processing, and write access for storing cache files in a .cache directory.
- Sanitization: No specific content sanitization or filtering is implemented in the provided template.
Audit Metadata