intake-chatbot
Pass
Audited by Gen Agent Trust Hub on May 12, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: Technical analysis of the skill instructions and associated reference materials shows a consistent and safe implementation for field data capture. The logic is focused entirely on construction industry terminology and reporting standards.
- [DATA_EXFILTRATION]: The skill is safe from data exfiltration as it lacks any network-enabled commands or tools. All operations are confined to the local filesystem, specifically managing project-related files like daily-report-intake.json and project-config.json.
- [COMMAND_EXECUTION]: There are no shell commands, script executions, or system-level interactions present. The skill logic is implemented through high-level AI instructions for data classification and entity resolution.
- [PROMPT_INJECTION]: The instructions do not contain any bypass, override, or developer-mode triggers. All classification rules are functional and do not attempt to manipulate the agent's safety guardrails.
- [SAFE]: Indirect Prompt Injection Surface Evaluation (Category 8): * Ingestion points: The skill processes conversational input provided by users via the /log trigger as specified in SKILL.md. * Boundary markers: There are no explicit delimiters defined in the instructions to separate user input from logic. * Capability inventory: The skill possesses the ability to read and write to local JSON project files; it has no network or shell execution capabilities. * Sanitization: The logic relies on strict NLP classification into a predefined JSON schema, which serves as an inherent structural filter for the ingested data.
Audit Metadata