Agent Guru
v1.0.0Use when building, designing, or reviewing a multi-agent system for production — routing agents, orchestrating subagents, guarding tools with permissions, ma...
⭐ 0· 36·0 current·0 all-time
byWeixuan Jiang@weixuanjiang
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
OpenClaw
Benign
medium confidencePurpose & Capability
The name/description (production multi-agent design) aligns with the content: detailed architecture patterns, tooling, and code examples for routing, orchestration, safety, memory, observability and persistence. It does not request unrelated credentials, binaries, or installs.
Instruction Scope
SKILL.md and the reference files contain runnable examples that read local files (e.g., AGENT.md from working_dir, ~/.agent, /etc/agent/global), connect to DBs/Redis/Postgres (example connection strings), spin up an HTTP endpoint, and fetch remote policy via httpx.get. Those are appropriate for the stated purpose (production agent harnesses) but they do instruct accessing filesystem and network resources — review and sandbox any copied examples before running.
Install Mechanism
Instruction-only skill with no install spec or shipped code — lowest install risk. Example pip install lines appear in docs (langgraph, langgraph-supervisor) but no code is downloaded by the skill itself.
Credentials
The skill does not declare required env vars or credentials, but examples reference environment-driven config (os.getenv), DB URLs, Redis/Postgres connection examples, and snapshotting of MAX_OUTPUT_TOKENS etc. These are reasonable for production guidance but you should not copy hardcoded credentials (e.g., 'postgresql://user:pass@db:5432/agents') into real deployments and should limit which env vars or secrets are used.
Persistence & Privilege
always is false and there is no install-time persistence or privileged modification of other skills. The guidance describes persistent components (checkpointers, vector stores) that are normal in production — the skill itself does not request permanent platform privileges.
Assessment
This is a content-rich, instruction-only playbook for production multi-agent systems — it appears coherent with that purpose. Before you copy or run any examples: (1) review and remove hardcoded credentials and replace with secured secrets; (2) sandbox code that reads files (AGENT.md, ~/.agent, /etc) to avoid unintentionally exposing local secrets; (3) validate any remote endpoints before allowing the agent to call them (the remote killswitch example calls an internal config URL); (4) adopt the GuardedToolNode / HITL patterns for any destructive tooling; and (5) if you need higher assurance, ask the publisher for provenance (homepage, repo) or run the code in an isolated dev environment. If you want a deeper risk review, provide the publisher/source URL or say which code snippets you intend to reuse.Like a lobster shell, security has layers — review code before you run it.
latestvk97frbj6q30hch50acd0j0v221840hgh
License
MIT-0
Free to use, modify, and redistribute. No attribution required.