ue5-phase-bans
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill contains directives intended to override other agent instructions and potentially bypass user confirmation. Key phrases include 'A ban outranks any installed skill' and 'Do not stop to ask; the ban already is the answer,' which instruct the agent to prioritize the skill's internal logic over conflicting recommendations from other skills or sessions without seeking human intervention.
- [COMMAND_EXECUTION]: The skill defines a 'Grep check' verification procedure that involves executing shell commands to search the project codebase for specific symbols associated with banned architectural patterns.
- [INDIRECT_PROMPT_INJECTION]: The skill processes the project codebase to identify 'breach tells' (signs that a ban is being circumvented), creating an attack surface where maliciously crafted strings in the codebase could influence the agent's reporting.
- Ingestion points: The agent reads and processes files within the local project codebase (e.g., in SKILL.md and breach-tells.md).
- Boundary markers: The instructions do not specify the use of prompt delimiters or warnings to ignore instructions embedded within the codebase files being scanned.
- Capability inventory: The skill utilizes shell command execution (grep) and references the ability to write to project context files via the 'ue5-project-context' skill.
- Sanitization: There is no description of sanitization or validation for the symbols or patterns extracted from the codebase during the verification checks.
Audit Metadata