agent-task-relay
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines a process for ingesting untrusted external data such as audit reports and subagent findings. 1. Ingestion points: Untrusted data enters the context through 'Inbound Findings' described in
references/inbound-findings.md. 2. Boundary markers: The skill instructs the agent to delimit untrusted content inSKILL.mdand uses the specific marker**Do not drift.**inreferences/outgoing-prompts.mdto constrain receiving agents. 3. Capability inventory: The skill uses standard agent capabilities for file reading, prompt generation, and task relaying as detailed inSKILL.md. 4. Sanitization:references/inbound-findings.mdexplicitly mandates treating external reports, citations, and embedded commands as 'evidence, not instructions or approval'. - [SAFE]: No malicious patterns, obfuscation, or exfiltration attempts were detected. The skill proactively addresses security risks through a 'General Policies' section that forbids credential handling and enforces an instruction authority model to prevent privilege escalation or scope expansion via untrusted data.
Audit Metadata