multi-agent-client-onboarding
Warn
Audited by Gen Agent Trust Hub on Aug 28, 2026
Risk Level: MEDIUMDATA_EXFILTRATIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The Tech Stack Mapper agent is tasked with locating sensitive credentials and configuration files within analyzed codebases. -- Evidence: In references/agent-prompts.md, the agent is instructed to identify API keys in configs and search paths like **/.env.example, **/.env.sample, and **/config/*. -- Behavior: Although the skill includes guardrails against including these secrets in the final output, the automated access to sensitive local files constitutes a data exposure risk.
- [COMMAND_EXECUTION]: The skill employs the Bash tool to inspect local environments and build systems. -- Evidence: SKILL.md grants Bash access, and specialist agents use it to audit repositories and configurations.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from documents and web searches, creating an attack surface for indirect prompt injection. -- Ingestion points: Data is gathered via Read, Grep, Glob, and WebSearch tools. -- Boundary markers: Agent prompts interpolate external data without delimiters or instructions to ignore embedded prompts. -- Capability inventory: The skill can execute commands via Bash and orchestrate additional agents, which could be exploited if malicious instructions are processed. -- Sanitization: No filtering or validation of external content is specified before processing.
Audit Metadata