agent-prompt-builder

Pass

Audited by Gen Agent Trust Hub on Jun 23, 2026

Risk Level: SAFEDATA_EXFILTRATIONEXTERNAL_DOWNLOADSPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill transmits agent observations and identity data to the Live Neon platform (agentpersona.live). This is a core functional requirement of the service and occurs over HTTPS to the vendor's own infrastructure.
  • [EXTERNAL_DOWNLOADS]: The skill facilitates the import of content from various external platforms, including GitHub, Twitter, and LinkedIn. This data is used to inform the agent's identity generation process.
  • [PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection, as it processes untrusted data from external sources to generate system instructions. It mitigates this risk by providing a manual review and approval mechanism (/prompt review) for all extracted beliefs and responsibilities.
  • Ingestion points: GitHub repositories, website content, Twitter feeds, RSS, and LinkedIn exports (as described in the sync and discover commands in SKILL.md).
  • Boundary markers: The /prompt review command provides a manual curation step before content is included in the prompt.
  • Capability inventory: The skill primarily uses curl to interact with the platform API; the generated prompt is intended for use in separate LLM calls.
  • Sanitization: The platform uses a multi-stage distillation process involving LLMs to extract signals from raw data.
  • [COMMAND_EXECUTION]: The skill utilizes curl and jq for API interactions and recommends the use of scheduling tools like cron to automate the identity evolution loop.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 23, 2026, 07:09 AM
Security Audit — agent-trust-hub — agent-prompt-builder