Yige
d52e61e041
## 摘要 / Summary ### 中文 四个智能体管理技能（create/update/delete/discover-agent）改用 `ManageAgent` 内置工具后，正文与工具契约大量重叠——ManageAgent 的 description + params 已**常驻每次 query 的上下文**，声明了五个 action 语义、参数约束、权限硬边界、错误语义、字段级合并、确认行为、list/get 返回格式；技能正文里再复述即冗余。本次对四技能做**强改写·语境融合**压缩： - 与工具契约重复的说明（参数/权限/错误码/确认行为/成功返回话术）**改写融入对应流程步骤**（如错误处理表 → 阶段一句、确认行为 → 阶段一句），不再照抄、不再表格化。 - 装饰性 ASCII 流程框图 → 一行文字流程；update 内部两张重复更新表 → 合并；YAML 元数据块（diff_metadata/context_handoff）融入流程；冗长示例（create 三份 JSON、update 附录 4 示例、discover ASCII 卡片）就地精简为代表示意。 - **不外置 references、不净删除任何内容**：所有决策/领域/交互意图完整保留——领域匹配表、persona/principles 的 L0/L1/L2 生成规范、update 两路径分流与字段级合并 vs 整体替换、防幻觉改名、回滚流程、discover 需求维度与无匹配衔接 create 等核心一字未丢。 zh 正文合计 **26542 → 9196 字符（降 ~65%）**；中英双份同步改写、逐段对齐，重算 i18n source_hash（validate-i18n 通过）。版本 create 2.5.1 / update 3.1.2 / delete 2.5.1 / discover 2.6.1，manifest 1.2.12。 ### English After the four agent-management skills adopted the `ManageAgent` builtin tool, their bodies heavily duplicated the tool contract — ManageAgent's description + params are **resident in every query's context** (action semantics, param constraints, permission hard-boundaries, error semantics, field-level merge, confirmation behavior, list/get return formats). This PR compresses all four via **aggressive rewrite + contextual fusion**: contract-duplicating text is rewritten into the relevant flow steps (not copied, not tabularized), decorative ASCII flow boxes become one-line text, update's two duplicate tables are merged, YAML metadata blocks are folded in, and long examples are trimmed in place to representative sketches. **No references externalization, no net deletion** — every decision/domain/interaction intent is preserved (domain matching table, persona/principles L0/L1/L2 generation spec, update's two-path split and field-level-merge-vs-full-replace, anti-hallucination rename, rollback flow, discover's need dimensions and create hand-off). zh bodies total **26542 → 9196 chars (~65% down)**; both locales rewritten and aligned, i18n source hashes recomputed (validate-i18n passes). Versions bumped, manifest 1.2.12.
6.1 KiB
name, description, version, type, risk_level, status, disable-model-invocation, tags, metadata, market
| name | description | version | type | risk_level | status | disable-model-invocation | tags | metadata | market | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| discover-agent | 根据用户需求推荐最匹配的智能体，展示候选列表并引导选择。Use when 用户描述需求但不确定该找哪个智能体帮忙，或想浏览可用的智能体。 | 2.6.1 | procedural | low | enabled | true |
|
|
|
discover-agent skill
L0: One-Sentence Summary
Match and recommend the most suitable Agent among registered Agents based on the user's need description.
L1: Overview
Procedural skill: understand the need → retrieve with ManageAgent(action='list') → semantic match scoring → rank and present → guide selection; on no match, hand off to create-agent automatically. Its value is what the tool can't give: semantic need matching, candidate ranking and presentation, and the create hand-off on no match. list / get are read-only, approval-free.
L2: Detailed Spec
Flow: need understanding → retrieval → match evaluation → ranking → presentation → guided selection.
Stage 1: Need Understanding
Trigger (any): user says "find me a… / is there a… / who can help me…", describes a task without naming an Agent, "which Agents are there", or the system detects a need that mismatches the current Agent. Extract dimensions from the description: domain (legal/finance/tech/education), task_type (consult/review/analyze/create), keywords (contract/report/code/paper…), urgency (routine/urgent).
Stage 2: Retrieval
ManageAgent(action='list') fetches all registered Agents (returns a compact list with name / id / status / description). Filtering: by default show non-offline Agents; offline ones appear only as a fallback when there's no better candidate; exclude internal system Agents (e.g. DesireCore itself) unless the user explicitly asks.
Stage 3: Match Evaluation
Judge match with LLM semantic understanding (not a formula): relevance of description / persona to the need, association of skills to the task type, domain fit, status availability (online preferred). Presentation tiers: strong match → mark "recommended", partial → "possibly relevant", no clear relation → don't show.
Stage 4: Ranking
Descending by overall score; ties broken by online status; show at most 5 candidates.
Stage 5: Presentation
- With matches: list candidates, each with name, description, key skills, status, and match score; ask the user to choose or refine the need. E.g. "1. Legal Advisor (92%) — contract review and legal risk assessment; skills: contract review / risk assessment / legal research; online".
- No match: report that none were found and offer three options — retry with a more specific description / create a new specialized Agent (hand off to create-agent) / browse all.
- Browse mode (user wants to see all): list name + description grouped by online / offline, and ask whether to view details of any.
Stage 6: Guided Selection
- Chose an Agent → switch to that Agent's conversation, passing the user's need context (source / target / user_intent).
- Wants more detail →
ManageAgent(action='get', id)for details (name / description / status / version / skill count / tool count / Git status); present the key info in natural language or a table and ask whether to chat. - Unhappy with candidates → guide the user to refine the need or suggest creating a new Agent.
- Chose "create new" → invoke the create-agent skill, passing the gathered requirements.
Collaboration and Error Handling
- Collaboration: on no match, hand off to create-agent (pass the need as initial info); on a successful match, optionally create a task and assign it to the target Agent.
- Errors: tool call fails → report error and suggest retry; empty Agent list → guide the user to create the first Agent; overly vague need → ask follow-ups and offer domain options;
getreturns "Agent not found: " → fall back tolistto reconfirm available Agents; recommended Agent in an abnormal state → mark the status and suggest picking an online one. list/getare read-only, approval-free, and risk-free; always done viaManageAgent.