| name | agent-skill-load-events |
| description | Track which skills are loaded during agent invocations. Use when analyzing skill usage patterns, identifying dead skills, or optimizing context loading. |
| license | MIT |
| compatibility | Claude Code, GitHub Copilot, OpenCode, Cursor, Codex, Gemini CLI |
| metadata | {"author":"paruff","suite":"uFawkesAI"} |
Skill: Agent Skill Load Events
Load trigger: "load agent-skill-load-events skill" > DORA: Cap 3 (Context Engineering)
Token cost: Low
Purpose
Track which skills are loaded during agent invocations to identify unused, overused, or inefficient skills.
Span Specification
agent.skill.loaded
Emit each time a skill is loaded for context.
Attributes:
| Attribute | Type | Description |
|---|
skill.name | string | Skill identifier (e.g. security/rbac-validation) |
skill.domain | string | Skill domain (e.g. security) |
agent.name | string | Agent that loaded this skill |
session_id | string | Unique invocation identifier |
Usage
Skills loaded 0 times in 90 days → archive candidates.
Skills loaded in every invocation → consider promoting to always-on context (but watch token budget).
Integration with Phase 0
The skills_loaded array in the Phase 0 invocation log is the source of truth.
Each entry in that array corresponds to one agent.skill.loaded span.
