Split skill usage from skill surfacing in Skills Graph

[antejavor]

Summary

Skills Graph currently records skills returned by list/search operations through the same usage relationship as skills that were actually retrieved or read from a valid local Skill Definition.

We have resolved the domain distinction:

• Skill Usage: a session used a skill during its work, such as through get_skill or reading a valid SKILL.md Skill Definition.
• Skill Surfacing: a skill appeared in a list/search result without proving the agent used it.

Problem

Today, surfaced skills can be recorded through the same USED_SKILL relationship with actions such as list_skills_result. That weakens future recommendation and evolution signals because being shown to an agent is not the same as being used by an agent.

Proposed direction

Introduce separate graph semantics:

(:Session)-[:USED_SKILL]->(:Skill)
(:Session)-[:SURFACED_SKILL]->(:Skill)

Keep USED_SKILL for actual use:

• get_skill
• valid local SKILL.md reads
• other explicit skill retrieval/application events

Use SURFACED_SKILL for weaker exposure signals:

• list_skills results
• search_skills results
• search_by_name results

Notes

This follows the context docs in context-graph/skills-graph/CONTEXT.md, where Skill Usage and Skill Surfacing are distinct terms.