| 1234567891011121314151617181920212223242526272829303132333435363738 |
- entity,field,type,nullable,status,group,description
- capability,id,VARCHAR PK,N,existing,标识,"格式 ""CAP-XXX""，跨 version 唯一"
- capability,name,VARCHAR,N,existing,标识,"能力名称"
- capability,version,VARCHAR(32),N,existing,标识,"多租户版本，DEFAULT 'v0'"
- capability,description,TEXT,N,existing,语义,"能力是什么（综合定位描述），用于人类阅读与 embedding 输入"
- capability,method,TEXT,N,new,语义,"去参数化的做法概述（核心机理层），用于跨 case 聚类与语义召回；具体参数下沉到 capability_tool.parameters_template"
- capability,category,TEXT,Y,new,组织,"树状分类，例 ""image/portrait/identity"""
- capability,effects,TEXT[],N,new,召回,"满足的原子需求描述列表，每条形如""实现 XX 效果""；需求-能力匹配的主信号（统一了原 applicable_to 概念）"
- capability,inputs,JSONB,N,new,执行契约,"输入契约：素材类型、参数 schema、约束；agent 调用前用于校验"
- capability,outputs,JSONB,N,new,执行契约,"产出契约：格式、规格、可校验属性"
- capability,preconditions,TEXT,Y,new,执行契约,"前置条件，例 ""需先有角色 LoRA""、""≥5 张参考图"""
- capability,criterion,TEXT,Y,existing,执行,"产出评估标准（提炼结果），用于质检和 agent 重试判断"
- capability,failure_modes,TEXT,Y,new,边界,"不适用的边界，反向过滤防误召回"
- capability,maturity,VARCHAR,Y,new,置信,"experimental / validated / production"
- capability,validation_count,INTEGER,Y,new,置信,"被多少 case 验证过"
- capability,last_verified_at,BIGINT,Y,new,置信,"最近一次验证时间戳（秒）"
- capability,embedding,float4[],N,modified,向量,"name + description + method + effects 拼接的向量（原为 name + description）"
- strategy,id,VARCHAR PK,N,existing,标识,"格式 ""strategy-XXX"""
- strategy,name,VARCHAR,N,existing,标识,"工序名称"
- strategy,version,VARCHAR(32),N,existing,标识,"多租户版本"
- strategy,description,TEXT,N,existing,语义,"工序概览描述"
- strategy,body,TEXT,Y,existing,语义,"人类可读的工作流叙述（设计审阅、故障排查入口）；与 steps 并存——人写 body 整理设计，agent 读 steps grounding"
- strategy,category,TEXT,Y,new,组织,"平台/介质/形态分类"
- strategy,effects,TEXT[],N,new,召回,"满足的复合需求描述列表（统一了原 goal_pattern + applicable_to）"
- strategy,inputs,JSONB,N,new,执行契约,"工序整体输入契约"
- strategy,outputs,JSONB,N,new,执行契约,"工序整体产出契约"
- strategy,preconditions,TEXT,Y,new,执行契约,"前置条件"
- strategy,steps,JSONB,N,new,执行,"结构化步骤数组：[{order, type: capability|sub_strategy|human_review|conditional, ref_id, inputs, outputs, params, quality_check, on_fail, on_success, optional}]；agent 执行 grounding 的权威源；与 strategy_capability/strategy_composition junction 同步写入"
- strategy,success_metrics,JSONB,Y,new,执行,"strategy 级整体成功判据"
- strategy,expected_cost,JSONB,Y,new,执行,"估算时长/费用，多候选排序用"
- strategy,coverage_estimate,FLOAT,Y,new,边界,"对需求的覆盖度估计（0-1，业务文档明确强调）"
- strategy,maturity,VARCHAR,Y,new,置信,"experimental / validated / production"
- strategy,validation_count,INTEGER,Y,new,置信,"被多少 case 验证过"
- strategy,last_verified_at,BIGINT,Y,new,置信,"最近一次验证时间戳"
- strategy,status,VARCHAR,N,existing,管理,"draft / approved / deprecated 等"
- strategy,created_at,BIGINT,N,existing,管理,"创建时间戳（秒）"
- strategy,updated_at,BIGINT,N,existing,管理,"更新时间戳（秒）"
- strategy,embedding,float4[],N,modified,向量,"name + description + effects 拼接的向量（原为 name + description）"
|
