The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the run_shell capability in its fallback path to execute a shell command via uvx for resource retrieval from a vendor package.
[EXTERNAL_DOWNLOADS]: The fallback path involves fetching the ouroboros-ai package from a public registry at runtime.
[DYNAMIC_EXECUTION]: The fallback routine uses a Python one-liner (python -c) within a shell command to load resources from the installed vendor package.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the user and the local codebase, which could contain malicious instructions targeting the agent's behavior.
Ingestion points: The interview topic (initial_context), repository-local files accessed via inspect_code, and external web data via web_research.
Boundary markers: The skill employs structured markers such as [from-user], [from-code], and [from-research] to maintain clear separation of data sources.
Capability inventory: The skill possesses capabilities for shell execution, subagent orchestration, and MCP tool calls.
Sanitization: The skill implements three distinct mandatory human-in-the-loop gates—the Refine gate, the Seed-ready Acceptance Guard, and the Restate gate—ensuring the user reviews and approves interpreted data before it affects code generation or final output.

interview