The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands to interact with the WeKnora API and an optional connection management tool.
Evidence includes the execution of curl for REST API operations and node scripts/consume-sse.mjs for processing streamed chat responses.
The skill also supports an optional arks CLI tool for managed connections using commands like arks invoke weknora.search --input request.json.
Mitigations: The instructions explicitly warn against passing API keys as literal command arguments or writing them to disk, favoring environment variable expansion instead.
[INDIRECT_PROMPT_INJECTION]: The skill processes content from an external knowledge base and chat sessions, which presents a surface for indirect prompt injection if those sources contain malicious instructions.
Ingestion points: Data enters the agent's context through the WeKnora REST API results (via curl) and the streamed output from scripts/consume-sse.mjs.
Boundary markers: The instructions encourage the agent to cite sources and avoid raw data dumps, although they do not mandate strict delimiter-based boundary markers for all external text processing.
Capability inventory: The skill has shell execution and network access to the user-provided WeKnora instance to fulfill its primary retrieval and ingestion functions.
Sanitization: The included consume-sse.mjs utility parses JSON and assembles the answer stream but does not perform semantic sanitization of the document content before it is processed by the agent.

weknora