Strong governance-focused design with 11 well-named tools covering identity, memory lifecycle, and audit. All tools have descriptions (avg 95 chars) and input schemas with enums. However, output schemas are not documented in the visible code, parameter descriptions lack format/constraint details (e.g., namespace format, valid_until ISO 8601 format not stated), and error handling guidance is absent. Tool names follow verb_noun pattern (register, authorize, write, boot, latest, read, transfer, revoke, expire, tier_upgrade, audit), excellent clarity. Schemas use Zod with proper types and enums. Missing: documented return types, actionable error messages, and per-parameter constraint documentation.

• Document return types for all tools. E.g., amh_read returns {memories: [{memory_id, content, created_at, source_tier, status}], total_count, next_cursor}. Include field types and descriptions.
• Add format constraints to parameter descriptions. E.g., 'namespace: Namespace identifier in format 'project:name' (alphanumeric + colon, 3-50 chars)' and 'valid_until: ISO 8601 datetime string (e.g., 2025-12-31T23:59:59Z)'.
• Implement confirmation pattern for irreversible ops. Add optional 'confirm' boolean param to amh_revoke, amh_expire, amh_transfer. When false, return {action, memory_id, current_state, would_change_to} without executing. When true, execute.
• Document error responses with recovery guidance. E.g., 'If amh_authorize returns permission_denied, the principal lacks write:namespace permission. Call amh_register_principal or request admin to grant access.'
• Clarify domain-specific terms in descriptions. Explain 'state markers' (e.g., '[state]', '[wrap-up]', '[handoff]' are special memory content prefixes), 'source_tier' levels (raw_source < llm_derived < human_confirmed), and when to use amh_boot vs amh_latest.