build-cognitive-execution-engine

Pass

Audited by Gen Agent Trust Hub on Aug 26, 2026

Risk Level: SAFEPROMPT_INJECTIONNO_CODE
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by interpolating potentially untrusted external data into subagent prompt templates.
    • Ingestion points: Untrusted data enters the agent context via placeholders such as {{TASK_DESCRIPTION}} and {{CONTEXT}} in implementer-prompt.md, and {{SPEC}} and {{IMPLEMENTATION_SUMMARY}} in spec-reviewer-prompt.md.
    • Boundary markers: The templates use Markdown headers and triple-dash delimiters (---) to separate sections, but they lack explicit escaping or 'ignore embedded instructions' shielding for the interpolated variables.
    • Capability inventory: Subagents are granted tool access via {{ALLOWED_TOOLS}} and are explicitly instructed to perform software implementation, which involves writing and verifying code.
    • Sanitization: There is no evidence of input validation, escaping, or filtering for the external content before it is interpolated into the prompts.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 26, 2026, 06:47 AM
Security Audit — agent-trust-hub — build-cognitive-execution-engine