deployment-pipeline-design
Pass
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill designs automation workflows that ingest untrusted data from external sources.
- Ingestion points: The pipelines ingest source code during the 'Source' stage and process external test results in the 'Integration Test' stage as described in
SKILL.md. - Boundary markers: The provided CI/CD templates (GitHub Actions, GitLab CI) lack specific delimiters or instructions to prevent the agent from executing instructions found within the processed code.
- Capability inventory: The defined stages include execution capabilities such as
docker build,kubectl apply, andnpm runwhich could be targeted by malicious payloads in the source data. - Sanitization: The examples do not demonstrate sanitization or filtering of inputs like environment variables or commit metadata.
- [COMMAND_EXECUTION]: The skill provides templates that execute shell commands for deployment and monitoring.
- Evidence:
SKILL.mdincludes commands forkubectl,docker, andcurlto manage infrastructure and verify health. - Context: These commands are standard for pipeline orchestration and are provided as design patterns for the user.
Audit Metadata