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

platform-pipeline-provider

platform-pipeline-provider

【开发与工程】Develop and debug LangBot platform adapters, HTTP Bot callbacks, message aggregation, pipeline stages, provider runners, model managers, local agent tools, and tool-loading workflows.

独立 GitHub 快照 · 自动分类 · 以原作者文档为准

技能说明

GitHub

name: platform-pipeline-provider description: "Develop and debug LangBot platform adapters, HTTP Bot callbacks, message aggregation, pipeline stages, provider runners, model managers, local agent tools, and tool-loading workflows." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

Platform, Pipeline, and Provider

Use this sub-skill for message flow, IM platform adapters, HTTP Bot and WebSocket chat behavior, pipeline stages, providers/runners, model manager, local agent, and tool manager issues.

Read First

Common Workflows

Trace one message

  1. Platform adapter converts native event into shared message/event entities.
  2. RuntimeBot applies routing rules and pushes messages to aggregation.
  3. MessageAggregator collapses N-to-1 bursts by session.
  4. Controller schedules Query work subject to concurrency.
  5. RuntimePipeline executes configured stages.
  6. Chat stage invokes providers/tools/plugins and records session/history.
  7. Output stages send responses through the adapter.

Add or change an adapter

Keep vendor translation inside the adapter. Do not put LLM provider logic or pipeline business rules inside platform code. Add or update manifests, config schemas, icons/assets, i18n strings, docs, and focused adapter tests.

Debug provider/tool behavior

Start with fake provider/tool tests before using real API keys. If plugin, Box, native, or stdio MCP tools are involved, route deep runtime details to plugin-box-skills after confirming the provider/tool manager boundary.

Bundled Helper

Use scripts/http_bot_hmac_helper.py for self-contained HTTP Bot HMAC signing, verification, payload building, and optional signed POSTs. It adapts the public example client into a skill-owned helper and does not depend on the original checkout.

Focused Checks

python scripts/select_langbot_checks.py pipeline
uv run pytest tests/smoke/test_fake_message_flow.py -q --tb=short
uv run pytest tests/integration/pipeline/test_full_flow.py -q --tb=short
uv run pytest tests/unit_tests/platform/test_http_bot_tenancy.py -q --tb=short
uv run pytest tests/unit_tests/provider/test_tool_manager.py -q --tb=short
我的收藏