tutorial-selfloop
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from
output/TUTORIAL.mdandoutline/module_plan.ymlto perform quality checks. If these files contain malicious instructions, they could influence the agent's behavior when reviewing the resulting report. - Ingestion points:
output/TUTORIAL.mdandoutline/module_plan.yml(processed bytutorial_contract_issuesinscripts/run.py). - Boundary markers: Absent; the script does not wrap ingested content in delimiters or add "ignore instructions" warnings.
- Capability inventory: File system write access (
output/TUTORIAL_SELFLOOP_TODO.md). - Sanitization: Absent; the content is processed for teaching components without explicit escaping or filtering of potentially malicious prompts.
Audit Metadata