pipeline-security
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: Automated detection flagged an override marker due to the presence of the phrase 'ignore previous instructions' in SKILL.md. This is a benign false positive, as the phrase resides entirely within a defensive 'Prompt Injection Safety Notice' section instructing the agent to ignore adversarial instructions in analyzed files rather than executing them.\n- [INDIRECT_PROMPT_INJECTION]: The skill analyzes untrusted third-party files, establishing an indirect prompt injection risk surface. This is addressed via the following evidence chain:\n
- Ingestion points: CI/CD configuration files and repository settings are located using Glob and parsed using Read inside SKILL.md.\n
- Boundary markers: The 'Prompt Injection Safety Notice' section explicitly instructs the agent to treat any embedded commands as text data to be analyzed rather than guidelines to obey.\n
- Capability inventory: The skill is constrained by the frontmatter to a read-only toolkit consisting of Read, Grep, and Glob. It contains no capabilities for code execution, file modification, or network operations.\n
- Sanitization: Instructions require validation against a strict output schema and generic redaction of credentials, though no automatic programmatic parsing constraints are enforced on input text.
Audit Metadata