界面已本地化;技能介绍与说明文档按所选语言由 AI 自动翻译,代码、命令、路径和专业术语保留原文,以原作者文档为准。
探索技能
开发与工程 · SKILL.md

configuration

configuration

【开发与工程】Use this RAGs sub-skill to inspect, edit, update, delete, and troubleshoot generated RAG agent configuration and cache state.

独立 GitHub 快照 · 自动分类 · 以原作者文档为准

技能说明

GitHub

name: configuration description: "Use this RAGs sub-skill to inspect, edit, update, delete, and troubleshoot generated RAG agent configuration and cache state." disable-model-invocation: true metadata: disco-role: operating license: MIT

RAGs Configuration

Use this sub-skill when a RAGs agent already exists and the task is to inspect or change its settings, understand persisted cache state, recover from stale cache issues, or delete an agent safely.

Read This When

  • The user is on the RAG Config page or mentions Update Agent, Delete Agent, agent_id, top_k, chunk_size, embed_model, llm, or additional tools.
  • The sidebar shows an agent but loading, updating, or deleting it fails.
  • The task involves AgentCacheRegistry, ParamCache, agent_ids.json, cache.json, or vector-index storage.
  • The app was upgraded and old cache files appear to break launch.

For initial bot construction, read ../builder/SKILL.md. For asking questions to an existing bot, read ../chat/SKILL.md.

Core Workflow

  1. Confirm an agent is selected. The config page only exposes editable controls when the current state has an agent_builder and cache.

  2. Inspect current fields: agent ID, system prompt, loaded data summary, summarization flag, additional tools, top_k, chunk_size, embed model, and LLM.

  3. Before editing cache files directly, run the read-only inspector:

    python sub-skills/configuration/scripts/inspect_agent_cache.py --cache-dir cache/agents
    
  4. Use the app's Update Agent path for normal changes. It deletes the old cache entry, updates selected fields, reconstructs the agent, and saves the new cache.

  5. Use the app's delete route for unwanted agents. Manual deletion is a last resort for stale or corrupt cache directories.

Read references/configuration.md for field semantics and update/delete workflows. Read references/cache-registry.md before editing or inspecting persisted cache. Read references/troubleshooting.md for known failure modes.

Important Operating Facts

  • The cache root is conceptually cache/agents relative to a RAGs checkout.
  • AgentCacheRegistry.get_agent_ids() returns an empty list when agent_ids.json is absent.
  • ParamCache.save_to_disk requires a vector index and writes both serialized cache metadata and vector-index storage.
  • RAGAgentBuilder.update_agent deletes the old cache for the current agent ID, sets the requested fields, calls set_rag_params, updates tools when provided, and then calls create_agent.
  • Additional tools are currently limited to web_search.

Safety Boundaries

The bundled cache inspector is read-only. Do not delete cache directories or rewrite agent_ids.json unless the user explicitly asks for destructive recovery and understands that generated agents may need to be rebuilt.

我的收藏