The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external artifact data (such as code, documents, or API responses) which it then uses to drive automated verification probes.
Ingestion points: The skill reads content from artifact_text or file_path (using inspect_code).
Boundary markers: There are no explicit instructions for the agent to treat this content as untrusted or to ignore instructions embedded within the artifact.
Capability inventory: The skill utilizes run_shell (command execution), orchestrate_subagents (browser and computer use), and inspect_code (file access) to evaluate the artifact.
Sanitization: No sanitization or filtering of the artifact content is specified before it is passed to probes or sub-agents.
[COMMAND_EXECUTION]: The skill implements a 'process probe' that uses run_shell to execute commands, applications, or smoke tests provided by the artifact being evaluated. This allows for the execution of arbitrary code within the shell environment for the purpose of behavior verification.
[EXTERNAL_DOWNLOADS]: In scenarios where the Ouroboros MCP server is unavailable, the skill provides a fallback mechanism that uses the uvx tool to download and execute code from the ouroboros-ai package. This package is a vendor-specific resource used to fetch the judge logic.

qa