The Agent Skills Directory

[SAFE]: No malicious patterns or security risks were identified. The skill architecture emphasizes security best practices through its 'Verify before you trust' core principle and explicit opt-in invocation policy.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it processes untrusted data from external sources like support tickets and worker handoff files. Ingestion points: evals/files/support-tickets.csv and various handoff documents. Boundary markers: Uses structured Markdown headers and specific templates from handoff-format.md. Capability inventory: Access to spawn_agent, send_input, and codex exec. Sanitization: Implemented via multi-layered verification gates, citation validation, and dedicated verifier agents.
[DYNAMIC_EXECUTION]: The skill dynamically generates prompts for subagents and utilizes the codex exec tool for orchestration tasks, which is a documented and expected capability for the Codex platform.

waves-codex