[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-80932":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":16,"subscribersCount":16,"size":16,"stars1d":17,"stars7d":18,"stars30d":19,"stars90d":16,"forks30d":16,"starsTrendScore":20,"compositeScore":21,"rankGlobal":10,"rankLanguage":10,"license":10,"archived":22,"fork":22,"defaultBranch":23,"hasWiki":24,"hasPages":22,"topics":25,"createdAt":10,"pushedAt":10,"updatedAt":41,"readmeContent":42,"aiSummary":43,"trendingCount":16,"starSnapshotCount":16,"syncStatus":44,"lastSyncTime":45,"discoverSource":46},80932,"AgentLoom","linora-u\u002FAgentLoom","linora-u","Simple, flexible workflow orchestration for multi-agent AI apps, with YAML configuration, runtime safety, observability, and resume support.","https:\u002F\u002Fgithub.com\u002Flinora-u\u002FAgentLoom#readme",null,"Python",79,5,3,4,0,33,46,47,99,2.33,false,"main",true,[26,27,28,29,30,31,32,33,34,35,36,37,38,39,40],"agent-framework","ai-agent","automation","cicd","code-review","developer-tools","llm","mcp","multi-agent","open-source","orchestration","python","unit-testing","workflow-automation","yaml","2026-06-12 02:04:08","\u003Cdiv align=\"center\">\u003Csub>\nEnglish | \u003Ca href=\"docs\u002Fcn\u002FREADME.md\">简体中文\u003C\u002Fa>\n\u003C\u002Fsub>\u003C\u002Fdiv>\n\n\n\u003Ch1 align=\"center\">AgentLoom\u003C\u002Fh1>\n\n\u003Cp align=\"center\">\n  \u003Cstrong>Build complex multi-agent applications with simple configuration, minimal glue code, safe runtime controls, and first-class observability.\u003C\u002Fstrong>\n\u003C\u002Fp>\n\n\u003Cp align=\"center\">\n  \u003Cstrong>AgentLoom helps developers turn multi-agent workflows into runnable, observable, resumable, and controllable applications.\u003C\u002Fstrong>\n\u003C\u002Fp>\n\n\u003Cp align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Flinora-u\u002FAgentLoom\u002Factions\u002Fworkflows\u002Ftests.yml\">\u003Cimg alt=\"tests\" src=\"https:\u002F\u002Fgithub.com\u002Flinora-u\u002FAgentLoom\u002Factions\u002Fworkflows\u002Ftests.yml\u002Fbadge.svg\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fwww.python.org\u002Fdownloads\u002F\">\u003Cimg alt=\"python >=3.12\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fpython-%3E%3D3.12-3776AB?logo=python&logoColor=white\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Flinora-u\u002FAgentLoom\u002Freleases\u002Ftag\u002Fv1.0.1\">\u003Cimg alt=\"release v1.0.1\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Frelease-v1.0.1-007EC6\">\u003C\u002Fa>\n\u003C\u002Fp>\n\n---\n\n## Quick Start\n\nAgentLoom is designed to get you from repository clone to a real multi-agent application quickly.\n\n```bash\ngit clone \u003Crepo-url> AgentLoom\ncd AgentLoom\n\nuv sync\n# If PyPI is slow or unavailable on your network:\n# UV_DEFAULT_INDEX=https:\u002F\u002Fmirrors.aliyun.com\u002Fpypi\u002Fsimple uv sync\n\ncp config\u002Fllm.example.yaml config\u002Fllm.yaml\n# Edit config\u002Fllm.yaml and add your model credentials.\n\nuv run loom run applications\u002Fai_quality_analysis\u002Fworkflows\u002Fcode_review_agent.yaml\n```\n\nAfter the first run, you should see:\n\n- structured terminal logs with agent names, task IDs, step duration, and token usage;\n- archived run files under `.logs\u002F\u003Cagent>\u002F\u003Ctimestamp>\u002F`;\n- resumable task state when checkpointing is enabled;\n- optional visualization through `uv run loom ui`;\n- optional terminal monitoring through `uv run loom dashboard`.\n\n## What AgentLoom Provides\n\nAgentLoom already implements the runtime pieces needed to build and operate complex agent applications:\n\n| Area | Implemented Capabilities |\n|---|---|\n| Multi-agent app assembly | Supervisor \u002F Worker roles, YAML `workflow`, `worker_agents`, direct `loom run`, generated entry scripts with `loom create`, and Python embedding through `run_app()`. |\n| Agent-as-Tool | Workers export as callable tools through `agent_function_schema`, with generated function signatures, required-input validation, docstrings, string results, and `.batch(tasks)`. |\n| Execution modes | Structured `tool_call` mode for traceable orchestration and `code_act` mode for flexible code-execution tasks. |\n| Python pre\u002Fpost processing | Custom tool functions and wrappers for scanning, caching, loops, retries, validation, error isolation, progress persistence, and artifact writing. |\n| Batch concurrency | Worker `concurrency: auto` or fixed concurrency, `tool.batch(tasks)`, back-pressure, circuit breaker, progress callbacks, and per-call state isolation. |\n| Skills and Hooks | Reusable Skills, force-injected \u002F on-demand \u002F hidden modes, lifecycle Hooks for tools, tasks, sub-agents, sessions, compaction, setup, and config changes. |\n| Model and context control | Per-agent `model_type`, multiple LLM endpoints, parameter inheritance, retry behavior, prompt customization, and multi-layer context compression. |\n| Tool ecosystem | Built-in file, shell, search, code-edit, git, todo, skill-loading, local Python tools, and local Codex Exec tools, plus MCP client integration. |\n| Local Codex integration | Register local `codex exec` as a normal function tool, with multiple aliases, `fixed_args`, `sandbox` \u002F `search` pass-through, and local Codex login\u002Fpermission behavior. |\n| Code intelligence | LSP service management for definition, references, symbols, hover, and workspace symbols, with tree-sitter fallback. |\n| Runtime safety | Path boundaries, include\u002Fexclude rules, per-agent policies, shell command\u002Foperator allowlists, shell security checks, sandbox wrapping, and `shell_audit.log`. |\n| Long-task resilience | Checkpoint resume, heartbeat detection, conversation recovery, tool-call error recovery, file history, worker skip-on-resume, and task cleanup. |\n| Observability | Rich terminal logs, plain text file logs, per-step duration, cumulative\u002Fincremental token usage, task\u002Fsubtask\u002Fagent context, and run archiving under `.logs\u002F`. |\n| UI and monitoring | Web UI with SSE updates, topology graph, timeline replay, multi-run grouping, plus TUI dashboard for active and resumable tasks. |\n| Reference applications | `ai_quality_analysis`, `unit_test_studio`, `repo_map`, and `codex_exec_demo` demonstrate direct runs, custom Python entrypoints, strict pipelines, concurrent Worker analysis, and local Codex tool calls. |\n\n## Why AgentLoom\n\n### Build Complex Agent Apps With Less Code\n\nMany agent frameworks give you components. AgentLoom gives you an application shape.\n\nComplex agent applications often grow the same glue code again and again: Worker registration, parameter adapters, runtime entrypoints, pre\u002Fpost processing, batch execution, logging, tracing, and safety controls. That code is costly because it is rarely reusable when the next agent application has a different workflow.\n\nWorker Agents are exported as callable tools, and Supervisor Agents load and schedule them automatically. That means a complex workflow can be assembled from focused agents, a small amount of business glue code, and a direct runtime entrypoint.\n\nAgentLoom moves the reusable parts into YAML and the runtime. YAML describes the workflow, Worker list, tools, models, permissions, and skills; Workers become callable tools; Supervisors load them automatically; Python stays focused on the application-specific parts such as preprocessing, postprocessing, caching, loops, validation, and artifact writing. Skills and Hooks package lifecycle behavior so it can be reused across applications.\n\nYou can start with:\n\n- `uv run loom run \u003Cworkflow>` to run an application directly;\n- `uv run loom create \u003Cworkflow>` to generate a Python entry script;\n- `run_app(\"\u003Cworkflow>\")` to embed an AgentLoom app inside your own Python pipeline;\n- `tool.batch(tasks)` to call the same Worker Agent across many inputs in parallel.\n\nThis is the core development-cycle win: you spend less time building orchestration plumbing and more time shaping the actual agent roles, tools, and task flow.\n\n## How AgentLoom Is Different\n\n| Common Agent Framework Pattern | AgentLoom's Approach |\n|---|---|\n| Component-first: users assemble low-level primitives themselves. | Application-first: run, generate, embed, monitor, and resume multi-agent apps directly. |\n| Each complex app tends to grow a new layer of glue code. | Reusable orchestration, runtime controls, observability, and lifecycle extensions live in AgentLoom; app code handles business-specific differences. |\n| Safety, recovery, and observability are often added later. | Runtime guardrails, checkpoints, logs, audit trails, UI, and dashboard are built into the framework path. |\n| Logs mostly describe model calls and final output. | Logs are designed to debug agent application structure: task, subtask, agent, step, token, checkpoint, and topology. |\n| Good for quick demos, but production-like runs need extra infrastructure. | Built for long-running, unattended tasks that need progress, recovery, and post-run analysis. |\n\n## Build A multi-agent App With Less Code\n\nAgentLoom's core model is simple:\n\n```mermaid\nflowchart LR\n    User[\"Application \u002F CLI\"] --> Supervisor[\"Supervisor Agent\"]\n    Supervisor --> WorkerA[\"Worker Agent as Tool\"]\n    Supervisor --> WorkerB[\"Worker Agent as Tool\"]\n    Supervisor --> LocalTools[\"Local Tools\"]\n    Supervisor --> MCP[\"MCP Tools\"]\n    WorkerA --> Runtime[\"Runtime Controls\"]\n    WorkerB --> Runtime\n    Runtime --> Logs[\"Logs \u002F Checkpoints \u002F UI\"]\n```\n\nThe important part is the boundary:\n\n- a Worker Agent defines its purpose, tools, inputs, and output contract;\n- AgentLoom turns that Worker into a callable Python tool with a real function signature;\n- the Supervisor calls Workers like tools and coordinates the full task;\n- business Python can still wrap the workflow for deterministic steps such as scanning, caching, validation, argument parsing, and artifact writing.\n\n### Where The Glue Code Goes\n\n| Work That Usually Becomes Glue Code | AgentLoom Placement |\n|---|---|\n| Reusable agent orchestration | YAML `workflow` and `worker_agents`. |\n| Worker parameter adaptation | `agent_function_schema` and generated function signatures. |\n| Project-specific pre\u002Fpost processing | Python wrappers registered as tools. |\n| Cross-application lifecycle behavior | Skills and Hooks. |\n\nThe `repo_map` application demonstrates this hybrid style: deterministic Python scans and ranks a repository, then AgentLoom calls Worker Agents to analyze directories and produce architecture documentation.\n\n## Example Applications\n\nThe `applications\u002F` directory contains working apps that show how AgentLoom is intended to be used.\n\n| Application | What It Builds | What It Demonstrates |\n|---|---|---|\n| `ai_quality_analysis` | Multi-dimensional code review application. | 12 Worker Agents, staged review, direct `loom run`, long-running codebase analysis. |\n| `unit_test_studio` | Python pytest generation workflow. | Strict multi-step pipeline, custom entry script, function intake, scenario planning, test generation, refinement, delivery report. |\n| `repo_map` | Repository architecture map generator. | Deterministic Python preprocessing plus Agent-driven analysis, bottom-up directory processing, `tool.batch(tasks)`, progress persistence. |\n| `codex_exec_demo` | Local Codex Exec tool-call example. | Direct `loom run`, normal function tool registration, `fixed_args` for Codex parameters, structured `tool_call` sequencing. |\n\nRun the default code review example:\n\n```bash\nuv run loom run applications\u002Fai_quality_analysis\u002Fworkflows\u002Fcode_review_agent.yaml\n```\n\nRun Repo Map with a custom Python entrypoint:\n\n```bash\nuv run python applications\u002Frepo_map\u002Frepo_map_app.py \u002Fpath\u002Fto\u002Fproject \\\n  --output_dir \u002Ftmp\u002Frepo-map-output \\\n  --exclude_dirs vendor \\\n  --exclude_dirs build\n```\n\nGenerate pytest tests for selected functions:\n\n```bash\nuv run python applications\u002Funit_test_studio\u002Fstudio_runner.py \\\n  \u002Fpath\u002Fto\u002Fyour\u002Fproject \\\n  \"src\u002Futils.py:parse_config,src\u002Fcore.py:run_pipeline\" \\\n  --output_dir tests\u002Fgenerated\n```\n\nRun the local Codex Exec tool example:\n\n```bash\nuv run loom run applications\u002Fcodex_exec_demo\u002Fworkflows\u002Fuse_codex_exec_demo.yaml\n```\n\n## Core Concepts\n\n### Agent As Tool\n\nWorker Agents can be exported as callable tools. AgentLoom generates function signatures, validates required inputs, builds the task payload, and returns a string result. The same Worker can also expose `.batch(tasks)` for parallel execution.\n\n### Supervisor \u002F Worker\n\nA Supervisor Agent coordinates the task. Worker Agents specialize in one part of the job. This keeps responsibilities explicit and makes the workflow easier to debug.\n\n### Workflow\n\nA workflow describes what an agent should accomplish and how the Supervisor should use its Workers. AgentLoom uses the workflow to build the runtime task specification and keep long-running runs aligned.\n\n### Skills And Hooks\n\nSkills add reusable knowledge or behavior. Hooks attach logic around task lifecycle, sub-agent lifecycle, tool calls, and other runtime events. They can be used for validation, memory, policy enforcement, result transformation, and visualization collection.\n\n### Execution Modes\n\nAgentLoom supports structured tool calling for controlled orchestration and code execution mode for more flexible programming tasks. Each agent can choose the mode that fits its role.\n\n### Local Codex Tool\n\nAgentLoom includes `src.tools.codex.codex_tool.codex`, which exposes local\n`codex exec` as a normal function tool. Register it in Agent YAML with\n`module\u002Ffunction`; no dedicated `system.yaml` Codex section is required.\n\nUse `fixed_args` to lock inputs such as `prompt`, `cwd`, `sandbox`, and\n`search`. Fixed inputs are bound by the framework and removed from the\nLLM-visible tool schema; unfixed inputs remain available for the LLM to provide.\n`sandbox: \"\"` omits `--sandbox` and lets the local Codex CLI configuration decide\npermissions. `search: \"true\"` passes `--search`; search is off by default. Before\nrunning the tool, make sure `codex` is on `PATH` and `codex login status`\nsucceeds.\n\n## CLI Reference\n\n| Command | Purpose |\n|---|---|\n| `uv run loom run \u003Cworkflow>` | Run an AgentLoom application. |\n| `uv run loom create \u003Cworkflow>` | Generate a Python entry script for an application. |\n| `uv run loom ui` | Open the Web visualization panel. |\n| `uv run loom dashboard` | Open the terminal task dashboard. |\n| `uv run loom list-tasks` | List resumable checkpoint tasks. |\n| `uv run loom clean-tasks` | Clean old checkpoint data. |\n\n## Documentation\n\n| Document | Description |\n|---|---|\n| [Configuration Overview](docs\u002Fen\u002Fconfig-overview.md) | Configuration layers, merge rules, and model configuration isolation. |\n| [Agent Configuration](docs\u002Fen\u002Fagent_config.md) | Supervisor\u002FWorker fields, tools, model selection, workflows, and skills. |\n| [LLM Configuration](docs\u002Fen\u002Fllm_config.md) | Model types, provider settings, inheritance, retries, and prompt cache settings. |\n| [System Configuration](docs\u002Fen\u002Fsystem_config.md) | Runtime settings, permissions, logging, execution environments, and tools. |\n| [Skills Configuration](docs\u002Fen\u002Fskills_config.md) | Skill package format, loading, invocation controls, and built-in skills. |\n| [Hooks Reference](docs\u002Fen\u002Fhooks.md) | Lifecycle events, hook types, matching, and execution behavior. |\n| [Checkpoint Resume](docs\u002Fen\u002Fcheckpoint.md) | Checkpoint layout, resume behavior, and long-task recovery. |\n\n## Development Skills\n\nThe `tools\u002Fskills\u002F` directory provides optional development skills for AI coding assistants working on AgentLoom applications:\n\n| Skill | Purpose |\n|---|---|\n| `create-app` | Generate an AgentLoom application scaffold. |\n| `create-skill` | Create a custom AgentLoom skill. |\n| `workflow-review` | Review Agent\u002FTool boundaries, orchestration contracts, and resilience design. |\n| `shell-security` | Configure shell execution safety policies. |\n| `update-skills` | Keep development skills aligned with source and docs changes. |\n\nThese are development aids, not required runtime dependencies.\n\n## Support & Contributing\n\nAgentLoom is built as a practical framework for complex agent automation. Issues, feature ideas, docs improvements, and example applications are welcome.\n\n- Open an issue: [GitHub Issues](https:\u002F\u002Fgithub.com\u002Flinora-u\u002FAgentLoom\u002Fissues)\n- Contact: [raine_walker@163.com](mailto:raine_walker@163.com?subject=AgentLoom%20Collaboration)\n- If the project helps you, a GitHub star makes it easier for others to find.\n","AgentLoom 是一个用于构建多智能体应用程序的简单灵活的工作流编排工具，支持YAML配置、运行时安全控制、可观测性和任务恢复。项目核心功能包括通过YAML定义工作流、自动化的智能体角色分配（如监督者和工作者）、以及丰富的执行模式选择，例如可追溯的`tool_call`模式和灵活的代码执行`code_act`模式。此外，它还提供了批量并发处理能力与Python预处理\u002F后处理支持，确保了应用在复杂场景下的高效稳定运行。该工具非常适合需要快速搭建和管理多个AI代理协作的应用场景，如持续集成\u002F部署(CI\u002FCD)、代码审查、自动化测试等。",2,"2026-06-11 04:02:53","CREATED_QUERY"]