---

name: maintainer-apis description: "Maintain MetaGPT internal APIs for serialization, memory, experience pools, skill management, repository parsing, YAML config models, and focused diagnostics." disable-model-invocation: true metadata: disco-role: operating license: MIT

Maintainer APIs

Use this sub-skill when the user asks to fix or inspect MetaGPT internals: serialization/deserialization, Message/Task schema models, memory persistence, experience pools, skill management, repository symbol parsing, YAML config models, focused maintainer tests, or public API inventories.

Route Here For

User intent	Use this sub-skill for	Key reference
"Fix serialization" or "deserialize a Role/Message/Team"	BaseSerialization, SerializationMixin, Message.instruct_content, polymorphic fields, save/load compatibility	Serialization workflow
"Debug memory/storage"	Memory, LongTermMemory, MemoryStorage, BrainMemory, RoleZeroLongTermMemory, RAG-backed persistence boundaries	Memory workflow
"Inspect exp_pool" or "update exp_cache"	ExperienceManager, Experience, serializers, scorers, context builders, perfect judges, config flags	Experience pool workflow
"Update skill management"	SkillManager, Chroma-backed retrieval, skill add/delete/get behavior, optional dependency checks	Skill manager workflow
"List public symbols"	Safe AST-based class/function inventory without importing provider/config-heavy modules	Diagnostics workflow
"Run focused maintainer tests"	Targeted pytest commands, ignored-test strategy, optional provider/RAG skips, collection-error handling	Testing guide

First Moves

1. Classify the change as serialization/schema, memory, experience pool, skill manager, repo parser, config model, or diagnostics before editing.
2. Prefer direct model construction with explicit Config/LLMConfig values over Config.default() when a placeholder API key may be present.
3. Run a focused test file or directory; avoid the broad pytest suite unless the user explicitly wants coverage and optional dependencies are ready.
4. Use scripts/list_public_symbols.py for API inventories instead of the original shell helper; it parses Python AST and does not import MetaGPT modules.

Runtime References

• Workflows: focused maintainer workflows, commands, edit strategy, coverage/report guidance, and public-symbol diagnostics.
• API Reference: key classes/modules, config keys, serialization contracts, and common edit points.
• Testing: pytest.ini behavior, safe focused commands, optional/provider tests, fixtures, and skip decisions.
• Troubleshooting: Pydantic v2, placeholder config, pickle/json compatibility, memory/exp_pool storage, optional dependencies, and generated-file issues.

Boundaries

• User-facing MetaGPT CLI, project generation, incremental updates, and recovery workflows belong in software-company; return here only for serialization internals behind those workflows.
• Data Interpreter and RoleZero workflow behavior belongs in data-interpreter; return here for RoleZero memory internals or exp_pool serializer/context-builder maintenance.
• RAG, search, browser, vector stores, and tool registry usage belongs in rag-and-tools; return here only when memory or exp_pool failures surface through those optional backends.
• Provider-specific API calls, browsers, Android devices, downloads, and long LLM runs are prerequisites or skips, not safe maintainer smoke tests.

Evidence provenance distilled from MetaGPT serialization, memory, exp_pool, management, repo parser, YAML config, pytest configuration, maintainer tests, and docs diagnostic scripts. Repo-relative paths in references are provenance/edit-point names for a MetaGPT checkout, not external runtime dependencies.