The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is extracting content from external, untrusted document formats, which creates a potential surface for indirect prompt injection when the output is processed by an LLM.
Ingestion points: The extract and extract_batch functions described in SKILL.md and the various language-specific API references ingest data from URIs, local files, and raw bytes.
Boundary markers: While the skill allows for page markers via marker_format, it does not implement default security boundaries for the extracted text itself.
Capability inventory: The library can perform file system reads and network requests for document retrieval (URL ingestion mode).
Sanitization: The documentation does not specify internal sanitization of extracted strings, placing the responsibility on the agent or user implementation.
[COMMAND_EXECUTION]: The skill provides instructions for installing and running the Xberg CLI tool using standard package managers (brew, npm, cargo) and execution proxies (npx, uvx). All commands target official vendor packages.
[EXTERNAL_DOWNLOADS]: The documentation references downloading prebuilt binaries from the project's GitHub releases page and pre-trained ONNX models from HuggingFace for embedding generation. These references are consistent with the vendor's infrastructure and the skill's technical requirements.
[SAFE]: The skill utilizes Base64 encoding in the CLI examples (e.g., --config-json-base64) to facilitate configuration passing. Analysis of the provided examples shows they decode to benign JSON configuration objects.

xberg