harness-wizard

Pass

Audited by Gen Agent Trust Hub on Mar 15, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill presents an indirect prompt injection surface by ingesting untrusted data that influences agent actions.\n
  • Ingestion points: The 'Application Specification' collected in Phase 8 of the wizard (SKILL.md) and stored in Archon documents.\n
  • Boundary markers: The generated prompt templates in .harness/prompts/ do not include explicit delimiters or instructions to ignore instructions embedded within the specification.\n
  • Capability inventory: The generated agents possess capabilities for shell command execution (bash), filesystem access, and repository management via the GitHub MCP server.\n
  • Sanitization: No validation or sanitization is performed on the specification content before it is processed by the agents.\n- [COMMAND_EXECUTION]: The skill generates environment initialization scripts (init.sh and init.ps1) and configures a pipeline where the 'Coder' and 'Tester' agents execute shell commands. Although it includes a phase for users to define security constraints (denying sudo, rm -rf, and curl), the core functionality depends on executing arbitrary code within the project directory.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 15, 2026, 11:08 AM
Security Audit — agent-trust-hub — harness-wizard