kb
Pass
Audited by Gen Agent Trust Hub on Aug 31, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill provides instructions to download and install the
kbCLI tool from the author's GitHub repository (github:hraness/kb#v0.17.0) using the Bun package manager. - [REMOTE_CODE_EXECUTION]: The skill automates tool installation by executing a shell script that downloads and installs a remote package if the binary is missing from the environment.
- [COMMAND_EXECUTION]: The skill interacts with the user's environment by running shell commands, which include capturing data from active browser sessions via the Chrome DevTools Protocol (CDP) debugging port and performing external lookups to the Archive.today service.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from URLs and PDFs, creating a surface for indirect prompt injection.
- Ingestion points: External data is retrieved using
kb clipandkb pdfcommands. - Boundary markers: The instructions include explicit warnings to the agent to treat ingested content as evidence and not as instructions.
- Capability inventory: The agent can execute CLI commands, perform network data retrieval, and write to the local filesystem.
- Sanitization: The skill uses metadata manifests to track data integrity and provides SDK utilities to safely package untrusted context.
Audit Metadata