The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill generates Python importer scripts and executes them locally using the kitaru importer test command to validate trace mapping. To mitigate risks, the instructions mandate running these tests in a credential-free isolated environment to prevent access to ambient project or cloud secrets, and a separate user approval checkpoint is required before executing any generated or untrusted code.
[EXTERNAL_DOWNLOADS]: The skill manages the installation of Python dependencies required for custom importers, potentially from external registries like PyPI. It mitigates supply chain risks by requiring exact pinned versions, provenance reviews, and explicit user authorization before installation.
[INDIRECT_PROMPT_INJECTION]: The skill processes external trace data (e.g., from Langfuse or Braintrust) which may contain malicious instructions intended to influence the agent during the import or mapping process. Defensive measures are implemented, including mandatory redaction of sensitive fields (prompts, completions, tool arguments), enforcing byte and structural depth limits, and requiring the use of safe deserializers like yaml.SafeLoader while explicitly banning unsafe ones like pickle.

kitaru-importer-builder