514-production-rollout-plan
Pass
Audited by Gen Agent Trust Hub on May 2, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes various CLI tools including the 514 CLI, git, and the GitHub CLI (gh) to manage deployments, environment variables, and source control. These operations are core to the skill's purpose of automating production rollouts within the vendor's ecosystem.
- [DATA_EXFILTRATION]: The skill accesses sensitive environment variables and deployment configurations using tools like
514 env getand514 env list. This access is confined to the vendor's infrastructure and is required for identifying target deployment resources during the planning stage. - [REMOTE_CODE_EXECUTION]: The skill has the capability to run arbitrary SQL queries via
514 clickhouse query. To mitigate the risk of unintended execution, the skill explicitly mandates an approval gate usingAskUserQuestionto confirm the SQL content with the user before execution. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted local data from files like
migrations/plan.yamlandmigrations/remote_state.json. - Ingestion points: Reads migration artifacts from the local
migrations/directory. - Boundary markers: Utilizes
AskUserQuestionto confirm the plan classification and any destructive operations with the user. - Capability inventory: Uses
Bashfor git, gh, and 514 CLI tool execution. - Sanitization: Relies on human-in-the-loop validation, requiring users to review the plan and the PR body before artifacts are committed or pushed.
Audit Metadata