weave-instrument
Pass
Audited by Gen Agent Trust Hub on Sep 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill is a legitimate developer utility for Weights & Biases instrumentation. It enforces strict credential safety by repeatedly instructing the agent to never read, print, copy, or commit API keys or secrets. It utilizes a 'login gate' mechanism in the SKILL.md file that halts execution if the user has not verified their identity via a local terminal, preventing accidental data leaks to unauthorized projects. All network endpoints target the official 'wandb.ai' domain.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to instrument LLM application traces, which naturally ingest untrusted user and tool data.
- Ingestion points: Data enters the tracing context through instrumented model calls, tools, and agent turns as defined in the SKILL.md and SDK references.
- Boundary markers: The instructions in references/trace_fidelity.md explicitly state to 'respect content suppression/redaction'.
- Capability inventory: The instrumentation involves network operations to 'trace.wandb.ai' and file/code modification to wrap calls using the 'weave' SDK.
- Sanitization: The skill advises on using synthetic data for tests and preserving structured outputs while respecting privacy redactions, which are industry-standard safety practices.
Audit Metadata