agentscope
AgentScope repo skill for building agents, provider connectors, RAG and memory workflows, service deployments, and local or sandboxed workspaces.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- agentscope
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- AgentScope repo skill for building agents, provider connectors, RAG and memory workflows, service deployments, and local or sandboxed workspaces.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/repo-provenance.md1.3 KB
- 参考文档references/repo-routing-metadata.json493 B
- 参考文档references/troubleshooting.md3.4 KB
- 脚本scripts/check_env.py3.3 KB
03 SKILL.md 正文
AgentScope
Use this skill for AgentScope agent workflows, provider setup, retrieval and memory, FastAPI service deployment, or workspace/sandbox execution.
Quick start
- Install the broad runtime set with
uv pip install "agentscope[full]"or install only the extras needed for one sub-skill. - Target Python 3.11 or newer.
- After install, run
python scripts/check_env.py --show-backendsfrom the skill tree to confirm imports and local backend availability. - Read
references/repo-provenance.mdwhen you need to check whether this skill matches the current checkout. - Read
references/troubleshooting.mdwhen install or import behavior is off.
Route map
sub-skills/agent-core/
Use for Agent, Toolkit, built-in tools, permissions, message/event handling, task tools, MCP tool wiring, and local skill loading.
Typical triggers:
- "How do I build an AgentScope agent?"
- "Why is this tool blocked or inactive?"
- "How do local skills load into a toolkit?"
sub-skills/provider-connectors/
Use for chat, embedding, formatter, and TTS provider classes, provider credentials, and provider-specific extras or environment variables.
Typical triggers:
- "Which extra do I need for Gemini/Ollama/XAI?"
- "How do I configure embeddings or TTS?"
- "Why does a provider import fail?"
sub-skills/rag-memory/
Use for KnowledgeBase, RAG middleware, vector stores, filesystem memory, mem0, ReMe, and the RAG or memory examples.
Typical triggers:
- "How do I index documents and search them?"
- "How do I attach RAG to an agent?"
- "How do mem0 or ReMe memory modes work?"
sub-skills/service-platform/
Use for create_app, storage, message buses, channels, hubs, MCP, schedules, knowledge-base service, and service bootstrap/deployment.
Typical triggers:
- "How do I start the AgentScope service?"
- "How do I wire Redis or SQLite storage?"
- "How do channels, hubs, or MCP fit into the service?"
sub-skills/workspace-sandboxes/
Use for local workspaces and sandboxed backends such as Docker, Bubblewrap, Apple Container, E2B, Daytona, K8s, and OpenSandbox.
Typical triggers:
- "How do I create a workspace?"
- "Which backend do I need for this sandbox?"
- "Why did archive import or backend initialization fail?"
How to choose
- If the task is about agent orchestration, start with
agent-core. - If the task is about provider credentials or model classes, start with
provider-connectors. - If the task is about retrieval or long-term memory, start with
rag-memory. - If the task is about HTTP service deployment or platform routes, start with
service-platform. - If the task is about the execution sandbox or workspace backend, start with
workspace-sandboxes.
If a task spans more than one area, read the most operationally specific sub-skill first, then cross-link to the others.
Shared reference and diagnostics
references/repo-provenance.md— current repo snapshot and refresh baseline.references/repo-routing-metadata.json— machine-readable routing metadata for repo-skill discovery.references/troubleshooting.md— cross-cutting install/import and package-selection problems.scripts/check_env.py— import and backend availability check for this skill tree.
Minimal import check
After installation, a quick check should succeed:
python -c "import agentscope; from agentscope.agent import Agent; from agentscope.tool import Toolkit"
If that import fails, fix the environment before using any sub-skill.
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).