autoresearch-prompt

Pass

Audited by Gen Agent Trust Hub on Jul 15, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill architecture creates a surface for indirect prompt injection by interpolating untrusted data into its prompt templates.
  • Ingestion points: Untrusted data enters the agent context from the eval_set.jsonl file (which contains external newsletter content) and from standard input (stdin) when the classify command is invoked.
  • Boundary markers: The skill does not employ boundary markers, XML delimiters, or specific instructions to isolate interpolated fields from the prompt instructions. Values are directly substituted for {{key}} placeholders in prompt.md.
  • Capability inventory: The skill has the capability to perform network requests to the Anthropic API and processes JSON responses returned by the model.
  • Sanitization: There is no evidence of input validation, escaping, or sanitization in src/autoresearch_prompt/prompt_loader.py before the data is rendered into the final system and user messages.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 15, 2026, 01:44 AM
Security Audit — agent-trust-hub — autoresearch-prompt