Feat/agent runner plugin#82
Closed
huanghuoguoguo wants to merge 72 commits into
Closed
Conversation
huanghuoguoguo
force-pushed
the
feat/agent-runner-plugin
branch
from
7aa3464 to
528e46f
Compare
Phase 0 integration complete - LangBot + SDK + runner repo minimal loop verified. Key changes: - Add RUN_AGENT action forwarding from runtime to plugin process - Add AgentRunner to preinitialize_component_classes for initialization - Implement Protocol v1 entities: AgentRunContext, AgentRunResult, capabilities, permissions - Add resp_message_id field to Message class for LangBot integration - Fix AgentInput.message_chain to accept list type (matches LangBot data) - Add comprehensive tests for AgentRunner protocol Integration verified: plugin:langbot/local-agent/default returns [stub] Echo response. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…support - Refactor AgentRunAPIProxy to inherit from LangBotAPIProxy, reducing code duplication - Pre-compute allowed resource IDs (frozenset) in __init__ for O(1) permission validation - Add STATE_SCOPE_LITERAL type annotation in AgentRunResult.state_updated() - Add AgentRunState TypedDict with 4 scopes (conversation, actor, subject, runner) - Remove unused base64 import from agent_run_api.py - Update documentation for params and state protocol Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
- Improve AgentRunAPIProxy with state scope support (conversation/session/run) - Enhance plugin handler for agent run operations - Add better error handling and logging
Add record of all 7 runner plugins migration completed on 2026-05-13
- Add INVOKE_RERANK to PluginToRuntimeAction enum - Add invoke_rerank() method to AgentRunAPIProxy - Support reranking documents with relevance scores
Reduces code duplication by composing LangBotAPIProxy instead of directly calling plugin_runtime_handler. Adds get_tool_detail() for LLM function calling support. Changes: - Compose LangBotAPIProxy for delegated API calls - Add permission validation before each delegation - Add get_tool_detail() method for tool schema fetching - Reduce from ~460 lines to ~350 lines
Define Protocol v1 AgentRunContext with required fields: event, delivery, context (ContextAccess), bootstrap, compatibility, metadata. Add history and event pull APIs to AgentRunAPIProxy with run_id authorization. Key changes: - AgentRunContext: Protocol v1 structure with required event/delivery fields - AgentEventContext: event_id, event_type, source required - DeliveryContext: surface, streaming/edit/reaction capabilities - ContextAccess: inline_policy, available_apis for pull APIs - BootstrapContext: optional bootstrap messages (NOT core history) - CompatibilityContext: legacy Query/Pipeline fields for migration - AgentRunResult factories: all require run_id parameter - AgentRunAPIProxy: history_page, history_search, event_get, event_page - PluginToRuntimeAction: HISTORY_PAGE, HISTORY_SEARCH, EVENT_GET, EVENT_PAGE - Legacy helpers: to_v1_result(run_id), create_legacy_context with event/delivery Runner docs updated: use ctx.run_id in factories, ctx.bootstrap.messages instead of ctx.messages (demoted to bootstrap for Protocol v1).
Return structured run.failed events for plugin-side RUN_AGENT failures, enforce plugin readiness before listing or running AgentRunners, and make AgentRunner class declarations flow into manifests.
Move the scoped type-check module list out of the test.yml workflow into a top-level mypy.ini ([mypy] files). The CI step becomes a generic `uv run mypy`, so the workflow no longer enumerates AgentRunner/MCP/daemon business modules; the scope is versioned with the source and reproducible locally with `uv run mypy`.
- ToolResource gains a parameters field (full JSON schema) so the host can prefill it and runners build LLM tools without get_tool_detail round-trips - langbot_list_assets exposes a 'skills' asset class so external harnesses can discover pipeline-visible skills under the unified tool model
huanghuoguoguo
force-pushed
the
feat/agent-runner-plugin
branch
from
e469927 to
c4a84f5
Compare
7 tasks
Member
|
continue in #88 |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
状态 / Status
QA 完成。 SDK 侧 Agent Runner plugin 支持已联调通过。本 PR 已包含 #2271 的 box 修复（#87 已合入
main，并随本分支 rebase 一起带入）。合并后需发布新 SDK 版本，供 LangBot 侧 bumplangbot-pluginpin。This PR carries the SDK-side AgentRunner Protocol v1 infrastructure and the skill-as-tool resource surface. QA complete; the #2271 box fix is included via rebase. After merge, a new SDK release is needed so LangBot can bump its
langbot-pluginpin.关联 PR / Related PR
main+ rebase 进本分支）: fix(box): recreate sandbox container when extra_mounts change #87关联测试仓库 / Related Testing Repositories
概述 / Overview
提供 AgentRunner Protocol v1 的 SDK 基础设施 + 技能即工具的资源面 / SDK infrastructure for AgentRunner Protocol v1 plus the skill-as-tool resource surface:
AgentRunContext:event/input/context/resources/runtime/delivery）、runtime dispatch、run-scoped Host 拉取 API（history / event / artifact / state）ToolResource.parameters（完整 JSON schema 预填，runner 免逐工具get_tool_detail）langbot_list_assets新增skills资产类（外部 harness 技能发现）extra_mounts变化则重建容器（#2271 修复，fix(box): recreate sandbox container when extra_mounts change #87）验收 / QA
scripts/use.py读到data/input.json、写回文件穿透回 host 技能 store，全exit 0完整验收矩阵见 LangBot 仓库：
skills/skills/langbot-testing/references/skill-all-tool-acceptance.md。