event-follow-up-engine
Pass
Audited by Gen Agent Trust Hub on Oct 25, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external sources, which introduces a potential attack surface.
- Ingestion points: It reads content from web pages via the fetch tool, uploaded requirements documents, attendee exports, and user-pasted text.
- Boundary markers: The skill contains high-quality defensive instructions requiring the agent to treat external content strictly as data. If an instruction is detected within this data, the skill is instructed to ignore it and report the attempt to the user at the beginning of the response.
- Capability inventory: The skill can read local files, execute a bundled validation script (scripts/text_check.py), and potentially read from CRM or marketing platform connectors.
- Sanitization: It uses a 'comprehension gate' that restates the user's request and requires a 'Yes' confirmation before final delivery.
- [COMMAND_EXECUTION]: The skill references and uses a local utility script,
scripts/text_check.py, for deterministic text verification. - Execution method: The agent is instructed to run this script to count characters, verify dates/weekdays, and detect restricted punctuation (e.g., em-dashes).
- Security Context: The script is a static file provided within the skill bundle and uses only standard Python libraries (datetime, os, re, sys). It does not download external code or execute dynamically generated strings.
- [SAFE]: The skill incorporates a 'Golden Rule' and 'Anti-Hallucination Rules' that prevent the AI from inventing data, benchmarks, or scoring weights. It also includes a proactive 'Scam Detection' mechanism that warns users if ingested files contain requests for sensitive information like bank details or government IDs.
Audit Metadata