github-actions-local-repro
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from GitHub workflow files and CI logs during the failure reproduction process. It lacks explicit boundary markers or sanitization to protect against instructions embedded within these external sources. The evidence chain includes ingestion from repository files (SKILL.md workflow steps), a lack of sanitization, and a capability inventory that includes shell execution of CI tools.
- [DYNAMIC_EXECUTION]: The skill constructs shell commands for act and action-validator by interpolating job names, event types, and workflow paths derived from local files. This dynamic command assembly creates a potential surface for command injection if the identifiers in the workflow files contain malicious shell characters or metacharacters.
Audit Metadata