skills/fw-ai/cookbook/debug/Gen Agent Trust Hub

debug

Pass

Audited by Gen Agent Trust Hub on Sep 9, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses shell commands to manage environment variables (FIREWORKS_SESSION_ID, FIREWORKS_CLIENT_SOURCE) and utilizes the firectl CLI tool to query job statuses. These operations are restricted to triage and diagnostic purposes within the vendor's ecosystem.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data such as job logs, error messages, and status outputs from the Fireworks API (referenced in references/triage-paths.md). This constitutes an ingestion surface where external content could theoretically influence the agent's behavior. However, the skill explicitly mandates a read-only approach and requires the redaction of customer data and credentials before any escalation, mitigating potential risks.
  • [SAFE]: The skill includes privacy safeguards, such as a telemetry notice and specific rules against sending raw errors or customer prose as journey telemetry. These design choices prioritize data protection during the debugging process.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 9, 2026, 05:59 PM
Security Audit — agent-trust-hub — debug