Kairn defines 3 tools with visible schemas and descriptions. All tools have input schemas with typed parameters and reasonable descriptions (100-250 chars). However, several quality gaps prevent a higher score: (1) Tool names lack clear verb prefixes, 'kn_add', 'kn_connect', 'kn_judge' are prefixed with 'kn_' (namespace) rather than starting with action verbs like 'add_node', 'create_edge', 'record_judgment'. This makes intent less clear to LLMs. (2) Parameter descriptions are present but minimal, most lack examples of valid inputs, format constraints, or dependency hints. (3) Output schemas are not documented, callers cannot see what fields are returned or how to chain tools. (4) Error handling descriptions are absent, tools do not explain what to do if a node already exists, if an edge weight is invalid, or if a relation verb is unrecognized. (5) Security considerations are undocumented, kn_judge and kn_connect modify state but descriptions do not warn about irreversibility or permission requirements.

• Rename tools to start with action verbs: 'add_node', 'create_edge', 'record_judgment' instead of 'kn_add', 'kn_connect', 'kn_judge'. Keep 'kn' as a namespace prefix (optional) but lead with the verb.
• Document output schemas for all tools. For kn_add: return {node_id: string, namespace: string, created_at: ISO8601, tags?: string[]}. For kn_connect: return {edge_id: string, source_id: string, target_id: string, weight: float, created_at: ISO8601}.
• Expand parameter descriptions with format and constraint details. E.g., 'Node name (1-255 alphanumeric + underscore, must be unique within namespace)' and 'Relationship type (required for kn_connect; use only for system-generated edges; prefer kn_judge for LLM-asserted relations)'.
• Add error recovery guidance to all tool descriptions. E.g., 'If node already exists, kn_add returns error code DUPLICATE_NODE; try kn_connect to link to the existing node instead.' and 'kn_judge only accepts 5 canonical verbs; if your relation doesn't fit, use kn_connect with a custom edge_type.'
• Enumerate the 5 canonical verbs in kn_judge's description: 'One of: conflicts_with, supersedes, compatible, scoped, related. (For other relation types, use kn_connect.)'
• Document that kn_connect and kn_judge are idempotent or not. If calling kn_connect twice with the same source_id, target_id, edge_type creates two edges or updates one, state this clearly.
• Add a 'security' or 'permissions' note to state-modifying tools explaining what access levels are required and whether the action is logged.