current-account-dataset-review
Pass
Audited by Gen Agent Trust Hub on May 17, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [DATA_EXFILTRATION]: The skill documentation identifies sensitive environment variables (e.g., TIANGONG_LCA_API_KEY, TIANGONG_LCA_API_BASE_URL) and mandates that they be sourced from secure environments rather than being hardcoded or processed directly by the skill logic, which is a standard security best practice.
- [COMMAND_EXECUTION]: The Node.js script 'run-current-account-dataset-review.mjs' acts as a dispatcher for the 'tiangong' CLI. It uses a predefined mapping of actions to CLI commands, which helps ensure that only intended commands are executed with the provided arguments.
- [PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection through the processing of untrusted data.
- Ingestion points: The agent reads local dataset snapshots in JSONL format via the '--input' argument across multiple actions such as 'validate' and 'rewrite-references' (SKILL.md).
- Boundary markers: The instructions do not define specific delimiters or instructions to ignore potential instructions embedded within the dataset files.
- Capability inventory: The wrapper script executes various CLI operations that include file system writes via '--out-dir' and remote API interactions for saving drafts via 'runTiangongCommand' (scripts/run-current-account-dataset-review.mjs).
- Sanitization: There is no evidence of content sanitization or instruction filtering for the input data processed by the wrapper script.
Audit Metadata