Intelligence snippets

Plugins list snippet contributions on their manifest:

intelligence: {
  snippets: [{ id: 'gitlab-clone', relativePath: 'snippets/gitlab-clone.md' }],
}

At materialisation time the resolver copies each file into memory/snippets/<id>.md (or the repo/tenant equivalent) so agents can read_memory({ file: "snippets/gitlab-clone.md" }) without bloating the global system prompt.

Why snippets?

• Document identifier shapes (namespace/project vs owner/repo).
• Capture webhook headers / signing quirks for operators.
• Encode service-account mapping for multi-account tenants.

Keep snippets short and actionable — they load on demand and should not duplicate entire skills.

Skills vs snippets

Asset	Declared via	Loaded
Skill	intelligence.skills or .claude/skills	Skill tool / read_skill
Snippet	intelligence.snippets	read_memory({ file: "snippets/..." })

Skills teach how to work; snippets hold facts about a provider deployment.