elite-deploy
Pass
Audited by Gen Agent Trust Hub on Apr 13, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The AI CI automation section in 'ai-ci.md' describes an 'Auto-Fix' workflow that introduces an indirect prompt injection attack surface.\n
- Ingestion points: The process extracts failure logs from CI environments, which can contain output from tests or scripts influenced by external or untrusted data.\n
- Boundary markers: The workflow description does not specify the use of delimiters or instructions to ignore potential commands embedded within the logs.\n
- Capability inventory: The skill describes capabilities for generating code patches, applying them to the filesystem, and committing changes to PR branches.\n
- Sanitization: The skill suggests some guardrails like diff size limits and human review for logic changes, but lacks explicit sanitization of the input logs to prevent the LLM from executing instructions hidden in the error messages.\n- [COMMAND_EXECUTION]: The 'deployment-strategies.md' and 'ci-pipeline.md' files contain numerous shell command templates for deployment operations.\n
- Evidence includes patterns for 'ssh', 'rsync', 'docker run', and 'nginx -s reload' used for orchestrating server deployments.\n
- While documented as examples, these represent powerful command execution capabilities if implemented by the agent as instructed.
Audit Metadata