agent-instructions-evaluator
Pass
Audited by Gen Agent Trust Hub on Oct 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface for indirect prompt injection because its primary function involves processing untrusted external data, such as agent instruction sets, YAML configurations, and Python/JSON tool definitions.\n
- Ingestion points: In
SKILL.md, the agent is directed to process 'Raw system prompts', 'Instruction blocks', and 'watsonx Orchestrate native agent YAML' files. The utility scriptsscripts/extract_agent_info.pyandscripts/extract_tool_info.pyare used to read and parse this content.\n - Boundary markers: The instructions lack explicit delimiters or 'ignore-embedded-instruction' directives to prevent the agent from accidentally obeying commands embedded within the instructions or code it is analyzing.\n
- Capability inventory: The agent has the ability to execute Python scripts on the local file system and generate/save report files (
agent_prompt_achievability_report.mdandagent_prompt_achievability_report_harness.json), which could be misdirected if an injection occurs.\n - Sanitization: No sanitization or filtering is applied to the analyzed content before it is processed by the agent.
Audit Metadata