Mindreader defines 8 tools with complete JSON schemas and descriptions. All tools have input schemas with typed properties and descriptions. However, descriptions are terse (avg ~60 chars, well below the 194-char baseline), lack WHEN/WHY context, and omit error recovery guidance. Parameter descriptions are minimal. Output schemas are not documented. No tool annotations (readOnlyHint/destructiveHint) despite clear risk levels (READ_ONLY, WRITE, REVERSIBLE, IRREVERSIBLE). Error handling is present but not recovery-focused. Naming is clear and verb-driven (recall, write, revise, withdraw, judge, place, unify). Schemas are well-formed with required fields and type constraints.

• Expand tool descriptions to 100 - 200 chars. For 'write': 'Write new facts to the graph with optional time qualification. Use this to assert knowledge, insights, patterns, or signals. Facts are immutable once written; use revise() to update or withdraw() to hide. Requires scope (layer membership) and subject/predicate/object triple.'
• Add tool annotations to mcp.json: readOnlyHint for recall/recall_semantic, destructiveHint for unify, idempotentHint for judge/place (if true). Example: {"name": "write", "description": "...", "inputSchema": {...}, "destructiveHint": true}
• Document output schemas in tool descriptions or a separate schema section. E.g., 'Returns {fact_iri, subject_iri, predicate, object, spike, effective, layers}' for write().
• Expand parameter descriptions with constraints and examples. For 'scope': 'Layer membership array. Empty [] selects only global/unlayered records. Named scopes like ["user-context", "session-123"] see global records plus those in requested layers. Required.'
• Add error recovery guidance to write/revise/unify descriptions: 'If unify fails with "nodes not found", verify node IRIs via recall(). If write fails with "predicate not allowed", check system-owned predicates (e.g., rdf:type).'