Plexus defines 3 tools with strong descriptions (194 - 250 chars, well above baseline 194 avg) and detailed parameter guidance. save_entity and load_skill have comprehensive schemas with type definitions and constraints. However, list_skills lacks output schema documentation, and no tools declare tool annotations (readOnlyHint/destructiveHint). Error handling is present but generic ('list_skills_failed'). Naming is clear and verb-driven. Overall: solid definitions with gaps in output documentation and error recovery guidance.

• Document list_skills output schema explicitly: { skills: Array<{id, name, description, version, category, trigger_phrases, context}>, count: number, hint: string }.
• Add tool annotations to save_entity: destructiveHint=true (creates/modifies graph state). Add readOnlyHint=true to list_skills and load_skill.
• Enhance error messages with recovery guidance: 'scope_denied: You do not have access to context "X". Available contexts: [list]. Try again with a permitted context.' instead of bare 'scope_denied'.
• Add constraints to save_entity.related: minItems=0, items schema with required properties {to_id: string, relation: string}, and enum constraint on relation values (derived_from, triggered_by, supersedes, documents, implements, etc.).
• Document load_skill body field size limit and add guidance: 'Returns full Markdown body (typically <50KB). For very large skills, consider pagination or summary-only mode.'

Inferred effective spec: 2026-07-28+.

• Relies on Stateful initialize / Mcp-Session-Id (removed; protocol is stateless) - make each request self-contained