software-company
Use MetaGPT's core virtual software-company workflow from CLI or Python, including project generation, incremental updates, recovery, config initialization, and custom Role/Action/Team patterns.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- software-company
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Use MetaGPT's core virtual software-company workflow from CLI or Python, including project generation, incremental updates, recovery, config initialization, and custom Role/Action/Team patterns.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/api-reference.md8.1 KB
- 参考文档references/troubleshooting.md7.5 KB
- 参考文档references/workflows.md9.0 KB
- 脚本scripts/inspect_role_action.py12.0 KB
03 SKILL.md 正文
Software Company
Use this sub-skill when a user asks how to run MetaGPT as a virtual software company, generate a project from a one-line idea, continue an existing project, recover a serialized team, initialize configuration, choose CLI options, or build a small custom Action/Role/Team workflow.
Quick Routing
| User intent | Use this sub-skill for | Key reference |
|---|---|---|
| "Generate a 2048 game" or "build a repo from this idea" | metagpt CLI and metagpt.software_company.generate_repo project generation |
CLI and Python recipes |
| "Explain MetaGPT CLI options" | Option names, defaults, boolean spelling, and safe dry checks | CLI option guide |
| "Configure OpenAI/Claude/Ollama" | config2.yaml initialization, precedence, provider fields, and placeholder-key diagnosis |
Configuration workflow |
| "Continue this project" | --inc, --project-name, --project-path, and QA rewrite inputs |
Incremental updates |
| "Recover a previous run" | --recover-path / recover_path validation and serialized team storage |
Recovery workflow |
| "Build a custom agent/team" | Safe Action, Role, RoleReactMode, Message, and Team patterns |
Custom roles and teams |
Before Running LLM Workflows
- Confirm Python is
>=3.9,<3.12; Python 3.9 or 3.10 is the safest choice, and use the root MetaGPT installation/config guidance when present. - Confirm
metagpt --helpworks before any project-generation command. - Initialize config with
metagpt --init-config, then replace placeholder secrets in~/.metagpt/config2.yaml; do not paste API keys into chat or generated files. - Treat
metagpt "Create a 2048 game",generate_repo(...),Team.run(...), debate examples, and custom roles that call_aask(...)as LLM-spending workflows requiring working provider config. - Use
scripts/inspect_role_action.py --check-cli-help --jsonfor safe no-network inspection of installed APIs and CLI availability, then use root troubleshooting guidance for cross-cutting provider/config failures when present.
Runtime References
- Workflows: CLI/Python recipes, output layout, config, incremental/recovery usage, custom agent patterns, and safe validation.
- API Reference: important signatures, classes, parameters, defaults, and behavior notes.
- Troubleshooting: config placeholders, provider failures, dependency issues, Mermaid/browser problems, Typer/click help,
--inc,recover_path, and long outputs.
Boundaries
- For Data Interpreter roles,
DataAnalyst/DI-generated project workflows, and notebook-style data analysis, route to thedata-interpretersub-skill. - For RAG, search, browser, tool registry, or retrieval internals, route to the
rag-and-toolssub-skill. - For AFlow, SPO, game examples beyond the core workflow, Android assistants, and environment extensions, route to
extensions-and-environmentswhen that sub-skill is present. - For serialization internals, experiment pools, repository parser internals, and maintainer APIs, route to
maintainer-apiswhen that sub-skill is present.
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).