kb-import
Pass
Audited by Gen Agent Trust Hub on Apr 25, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill identifies and processes untrusted content from markdown files in the documentation directory to update the project's core configuration files.
- Ingestion points: Local markdown files within the
docs/kb/directory are read and parsed to extract tags, topics, and loading scopes. - Boundary markers: There are no explicit markers used to separate the content being read from the agent's internal instructions.
- Capability inventory: The skill is restricted to file system read and write operations on specific project files (
CLAUDE.md,_index.md,_log.md). It lacks network access or the ability to execute arbitrary shell commands. - Sanitization: A robust human-in-the-loop mechanism is present in Step 4, requiring the user to verify and approve the extracted data (Topic, Path, Scope, Tags) before the configuration is updated.
Audit Metadata