The Agent Skills Directory

[SAFE]: No malicious instructions, bypass attempts, or obfuscated content were detected in the skill orchestrator or its subagent scripts.
[SAFE]: The skill follows a structured, multi-step validation process that promotes accuracy and evidence-based reasoning, with built-in limits on execution loops.
[SAFE]: The architectural risk of indirect prompt injection is handled through structured subagent communication and limited tool capabilities. 1. Ingestion points: Untrusted data enters through the user request and web search results processed by the subagents. 2. Boundary markers: Absent, but mitigated by the requirement for structured reports from subagents which provides a degree of process isolation. 3. Capability inventory: Restricted to platform-native subagent and task tools; no access to shell commands, sensitive files, or direct network operations is present. 4. Sanitization: Not explicitly implemented, as the skill relies on model-based evaluation and platform guardrails to process findings.

recency-guard