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.
7.6 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 | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| create-agent | 通过多轮对话收集需求，调用 ManageAgent 内置工具创建新的 AgentFS v2 智能体，支持自定义 persona 和 principles。Use when 用户要求创建新智能体、培养某领域助手、或快速基于模板生成可治理 Agent。 | 2.5.1 | meta | low | enabled | true |
|
|
|
create-agent skill
L0: One-Sentence Summary
Collect requirements through natural-language conversation and call the ManageAgent builtin tool to create a specialized digital Agent.
L1: Overview
Meta-skill: gather requirements over multi-turn conversation → generate persona/principles → land via ManageAgent(action='create'). Use it to raise a domain specialist for the user (legal advisor, financial analyst, etc.). Its value is what the tool can't give: domain-tailored persona/principles generation + a pre-create preview confirmation.
L2: Detailed Spec
Flow: intent recognition → requirement gathering → content generation → user confirmation → creation → receipt.
Stage 1: Intent Recognition
Trigger (any): user explicitly says "create an Agent / make me an assistant"; describes a domain-specific need the current Agent lacks; asks "can you raise a …". Confirm intent, then move to requirement gathering.
Stage 2: Requirement Gathering
- Required:
name,role(core responsibility),target_users,domain(specialty). - Optional:
style(communication tone),boundaries(red lines),language(default Chinese) — defaults derived from the domain template. - Strategy: prefer inferring from the user's natural description; only ask for missing required fields; at most 2 questions per turn.
Stage 3: Content Generation
Organize persona and principles as structured fields (do not emit raw markdown; present field-by-field). Leave uncollected fields empty for the system to fill with defaults:
- persona: L0 one-sentence core identity; L1
role/personality(array of trait tags) /communication_style; L2 specialty, values, decision preferences (free-form). - principles: L0 one-sentence top principle; L1
must_do/must_not(safety red lines) /priority; L2 governance principles, escalation rules (free-form).
Domain matching reference (recommend personality and red lines by domain):
| Domain | Recommended personality | Default must_not |
|---|---|---|
| Legal / contract | Professional, rigorous, prudent | No litigation representation; not a substitute for formal legal advice |
| Finance / accounting / investment | Precise, analytical, conservative | No investment advice; no real transactions |
| Code / development / architecture | Logical, pragmatic, direct | No direct production access; no credential storage |
| General / other | Friendly, helpful | Standard safety norms |
Stage 4: User Confirmation
Present the preview in natural language / tables (name, description, persona, principles; no raw markdown source), e.g.:
About to create "Legal Advisor Assistant" — focused on contract review and legal risk assessment. Persona: digital legal advisor; professional, rigorous, prudent; uses legal terms accurately with plain-language explanations. Principles: user interest first, not a substitute for formal legal advice; must — cite statutes / mark uncertainty / recommend consulting a lawyer; must not — litigation representation / leaking consultations; priority: user safety > accuracy > efficiency. Confirm creation? (Confirm / Modify / Cancel)
If the user picks "Modify": ask which field → re-collect that field → update the preview → show and confirm again.
Stage 5: Create via ManageAgent
Structured call (persona/principles also accept markdown strings; unprovided fields are auto-filled):
ManageAgent({
"action": "create",
"name": "Legal Advisor Assistant",
"description": "A digital Agent focused on contract review and legal risk assessment",
"persona": { "L0": "…", "L1": { "role": "…", "personality": ["professional","rigorous","prudent"], "communication_style": "…" } },
"principles": { "L0": "…", "L1": { "must_do": ["…"], "must_not": ["…"], "priority": "user safety > accuracy > efficiency" } }
})
- Minimal create needs only
{ "action": "create", "name": "My Assistant" }(everything else auto-generated). - Optional
id(kebab-case slug;core/desirecoreare reserved core identifiers and cannot be used, including when the slug auto-generated fromnamecollides),config.llm(llm delta only; sensitive fields like mcp_servers are rejected and must be adjusted via the UI after creation). - The Agent is registered and usable immediately; the returned ID works directly with ManageTeam / Delegate. If the tool errors (already exists / reserved identifier / non-whitelisted config field, etc.), explain the reason and retry after adjusting per the hint.
Stage 6: Receipt
Present in a user-friendly way (no internal paths / technical details): confirm success and suggest next steps — start chatting, add skills, adjust persona or rules.
Background and Constraints
- AgentFS structure, troubleshooting, and protected paths: see
_agentfs-background.mdand_protected-paths.yaml. - Always create via
ManageAgent; never use curl / HTTP or write AgentFS directories directly. Requires client ≥ 10.0.90.