5. Knowledge base data lookups

[mocha06]

Fifth phase of the milestone. Depends on #413 (edits the same KB modules; merge serially).

Tools

MCP	CLI
get_ai_knowledge_base_data_lookup	pipefy kb data-lookup get
create_ai_knowledge_base_data_lookup	pipefy kb data-lookup create
update_ai_knowledge_base_data_lookup	pipefy kb data-lookup update
delete_ai_knowledge_base_data_lookup	pipefy kb data-lookup delete

Scope

• Typed condition models: every condition requires field + operator; AI-filled conditions (usingFillWithAi) additionally require inputName, inputType, inputDescription; static conditions require value. operator is an opaque string (documented examples only). outputFields capped at 30.
• sourceRepoId is treated as an opaque non-blank ID (no format coercion).
• Reads never return conditions (the API doesn't expose them), and updates replace the full definition — the update tool therefore requires the complete definition (sourceRepoId, searchQuery, outputFields, non-empty conditions) on every call, and docs tell callers to keep their lookup definition as the source of truth client-side.
• Deletes require confirmation; CLI writes gated on the access probe.

Acceptance

• Full SDK + MCP + CLI parity, docs/parity.md, skills updated (create with AI-filled condition → attach → update with full replacement).
• Tests cover condition typing rules, full-definition update enforcement, and confirm-delete.