instrument-existing-agent-with-prefactor-sdk
Pass
Audited by Gen Agent Trust Hub on Sep 22, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONCREDENTIALS_UNSAFE
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill references several Node.js packages under the vendor-controlled @prefactor namespace (e.g., @prefactor/ai, @prefactor/langchain, @prefactor/core) to be installed via standard package managers. These are official resources provided by the vendor for agent instrumentation.
- [COMMAND_EXECUTION]: Instructions include using the prefactor CLI for setup, resource registration, and telemetry verification, alongside standard package manager commands like bunx or npx. These operations are consistent with the skill's primary purpose of tool integration.
- [INDIRECT_PROMPT_INJECTION]: The skill facilitates the capture of agent inputs and outputs for tracing, which constitutes an ingestion surface.
- Ingestion points: Tracing occurs at agent boundaries, specifically LLM and tool call inputs/outputs as detailed in SKILL.md.
- Boundary markers: The skill explicitly advises using truncation and redaction for captured data.
- Capability inventory: The skill records metadata, token usage, and execution status for observability.
- Sanitization: The implementation rules mandate redaction of secrets and truncation of payloads to mitigate the risk of exposing sensitive or malicious content within traces.
- [CREDENTIALS_UNSAFE]: The skill mentions handling PREFACTOR_API_TOKEN and configuration files like prefactor.json. It proactively includes security best practices by instructing users to verify that these files are ignored by git to prevent accidental credential leakage.
Audit Metadata