add-agent-tool
Pass
Audited by Gen Agent Trust Hub on Feb 21, 2026
Risk Level: SAFE
Full Analysis
- [Prompt Injection] (SAFE): The instructions are purely technical and do not contain attempts to override agent behavior or bypass safety guardrails.
- [Data Exposure & Exfiltration] (SAFE): No hardcoded credentials or unauthorized network operations were detected. The file mentions standard workspace file operations expected for an IDE extension.
- [Obfuscation] (SAFE): No Base64, zero-width characters, or hidden payloads were found.
- [Unverifiable Dependencies] (SAFE): The skill does not perform any remote package installations or download external scripts.
- [Indirect Prompt Injection] (LOW): The skill describes the creation of tools that ingest user-provided data (e.g., file paths). While this defines a potential attack surface for the tools being built, the guide encourages the use of structured schemas and existing path resolution utilities to mitigate risks. Documented evidence:
- Ingestion points: Tool parameters (params.path, params.file) in
src/agent/tools/. - Boundary markers: Not explicitly detailed in the documentation snippets.
- Capability inventory: File reads, writes, terminal command execution, and file edits.
- Sanitization: Recommends using the shared
resolveWorkspacePathutility for path validation.
Audit Metadata