The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted trace data (inputs and outputs) from the Opik platform to generate test assertions.
Ingestion points: The skill retrieves trace content, specifically trace.input and trace.output, using the opik SDK (SKILL.md, Step 2).
Boundary markers: There are no explicit prompt delimiters or "ignore embedded instructions" warnings applied when the trace data is stored as a test suite item.
Capability inventory: The skill is permitted to use Bash, Read, Grep, and Glob tools for local file and project interaction.
Sanitization: No explicit sanitization or escaping of the ingested trace content is described before it is written back to the Opik test suite.
[EXTERNAL_DOWNLOADS]: The skill utilizes the opik Python SDK to interact with the Opik service. This is an official library provided by the vendor (Comet ML).
[DATA_EXPOSURE]: The skill checks for the existence of ~/.opik.config and utilizes the OPIK_API_KEY environment variable. This is standard and expected behavior for managing credentials and configuration for the Opik platform.
[COMMAND_EXECUTION]: The skill is authorized to use the Bash tool to perform repository-level tasks and interact with the Opik SDK or MCP tools as part of its primary function.

opik-test