Agent Observability

Use this skill to make Skill runs observable without leaking sensitive data, while providing structured data for later analysis.

Contract

• Input: skill execution context and raw execution data
• Output: redacted execution record, structured analysis data, and traceability links
• Scope: observation of skill execution for quality improvement and debugging
• Rule: all personally identifiable information, secrets, and sensitive data must be redacted
• Rule: structured data must enable analysis of patterns without compromising privacy
• Rule: outputs must link to validation evidence when available

Steps

1. Capture Execution Context

• Capture the Skill name, version, context pack used, and tools invoked
• Record timestamp, duration, and compute resources used
• Note the inputs provided to the skill and the triggering context