meta-gpt
Use MetaGPT for multi-agent software-company workflows, Data Interpreter tasks, RAG/tools, extensions/environments, and repository maintenance.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- meta-gpt
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Use MetaGPT for multi-agent software-company workflows, Data Interpreter tasks, RAG/tools, extensions/environments, and repository maintenance.
- disable-model-invocation
- true
02 资源文件 5
- 参考文档references/installation-and-configuration.md4.2 KB
- 参考文档references/repo-provenance.md2.3 KB
- 参考文档references/repo-routing-metadata.json491 B
- 参考文档references/troubleshooting.md4.4 KB
- 脚本scripts/check_metagpt_environment.py5.8 KB
03 SKILL.md 正文
MetaGPT
Use this repo skill when the user asks about MetaGPT, the metagpt Python package, the metagpt CLI, multi-agent software-company generation, Data Interpreter, RoleZero agents, MetaGPT RAG/tools, AFlow/SPO/extension environments, or maintaining MetaGPT internals.
Start Here
- Read installation-and-configuration.md before running MetaGPT, changing providers, diagnosing config, or selecting optional dependencies.
- Run scripts/check_metagpt_environment.py for safe import/config/CLI diagnostics that do not call an LLM or write a project.
- Read troubleshooting.md when imports, config validation, provider calls, optional dependencies, browser/Mermaid, or CLI commands fail.
- Read repo-provenance.md before deciding whether this skill is current for a MetaGPT checkout.
Route By Task
| User task | Read |
|---|---|
Generate a project from a one-line idea, use metagpt CLI options, initialize config, continue an existing project, recover a team, or build custom Action/Role/Team workflows |
software-company |
| Use Data Interpreter, RoleZero, DataAnalyst, SWEAgent, notebook/code execution, DI benchmark prompts, or tool-assisted data tasks | data-interpreter |
| Build or debug RAG pipelines, document stores, retrievers/rankers, search/browser/data/editor tools, tool registry, or optional RAG dependencies | rag-and-tools |
| Use AFlow, SPO, Android Assistant, Stanford Town, werewolf/Minecraft/API environments, CR/SELA, or other extension integrations | extensions-and-environments |
| Maintain serialization, memory, experience pools, skill management, repo parsing, config models, focused tests, or public symbol inventories | maintainer-apis |
Install Baseline
MetaGPT is packaged as metagpt==1.0.0 and declares Python >=3.9,<3.12. Prefer Python 3.9 or 3.10 for production-style use when following the project docs; Python 3.11 is useful for inspection but some dependencies can be slow or version-sensitive.
python -m pip install metagpt
# or, for a local checkout
python -m pip install -e .
metagpt --help
Before any workflow that calls an LLM, create or fix ~/.metagpt/config2.yaml and replace placeholder API keys. ~/.metagpt/config2.yaml has higher precedence than a checkout-local config/config2.yaml.
Safe Validation
Use safe checks before expensive runs:
python scripts/check_metagpt_environment.py --json
metagpt --help
Do not use project generation, Data Interpreter, AFlow/SPO optimization, external environment simulations, browser automation, email automation, or RAG indexing as smoke tests unless the user has explicitly provided credentials, data, services, and cost approval.
Capability Map
- Software-company generation:
metagpt "Create a 2048 game",metagpt.software_company.generate_repo(...), incremental--incmode, team recovery, custom roles/actions. - Data Interpreter: planning, generated-code execution, data analysis, benchmark prompts, RoleZero-family roles, browser/OCR/email/Stable Diffusion examples behind explicit prerequisites.
- RAG and tools: document stores, retrievers/rankers, RAG factories, search providers, browser engines, tool registry/recommendation, optional dependency diagnostics.
- Extensions/environments: AFlow and SPO optimizers, Android Assistant, Stanford Town, werewolf, Minecraft/API/software environments, CR/SELA experiments.
- Maintainer APIs: serialization/deserialization, memory storage, experience pools, skill management, repo parser, config models, focused pytest selection.
Common Decisions
- If a task is about using MetaGPT as a package or CLI, route to
software-companyunless Data Interpreter, RAG/tools, or extensions are named. - If a task names a missing optional package, vector store, browser, Android device, dataset, or external service, route to the owning sub-skill and root troubleshooting before installing broad extras.
- If a task edits MetaGPT source code, route to
maintainer-apisfor focused tests and internals, then cross-link to the user-facing sub-skill that owns the affected workflow. - If a run would call an LLM, spend tokens, send data to a provider, browse the web, reply to email, download datasets, or run a simulation, ask for explicit credentials/safety confirmation first.
04 同仓库更多 Skills
passnet-triton-opt
Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.
passnet-solve
END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.
passnet-skill
PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.
passnet-pattern-fusion
Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.
passnet-orchestrate
Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.
passnet-feedback
Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).