agentsope/SkillAlchemy / skills/agentsop-crewai
agentsop-crewai
VerifiedSOP for building multi-agent systems with CrewAI — role-based collaboration, sequential/hierarchical processes, Flows, memory, delegation. Use when modeling agent teams with clear roles and task pipelines.
| File | Kind | Size |
|---|---|---|
| README.md | markdown | 1.8 KB |
| SKILL.md | skill md | 24.2 KB |
| intermediate/operation_candidates.json | config | 6.4 KB |
| references/R1-architecture.md | markdown | 4.8 KB |
| references/R2-sop-workflow.md | markdown | 5.3 KB |
| references/R3-dilemma-cases.md | markdown | 7.0 KB |
| references/R4-anti-patterns.md | markdown | 4.8 KB |
| references/R5-ecosystem-context.md | markdown | 5.4 KB |
intermediate/operation_candidates.json
View on GitHub{
"framework": "CrewAI",
"version_captured": "1.14.5 (2026-05-18)",
"capture_date": "2026-05-19",
"operations": [
{
"id": "OP-1",
"name": "Decide whether to use CrewAI",
"trigger": "User describes a task involving team/collaboration/roles or explicitly mentions CrewAI",
"action": "Run Phase 0 decision tree: (1) multi-role check; (2) control-flow check. Output: use-CrewAI-Sequential | use-CrewAI-Flow | switch-framework",
"output": "Framework decision + rationale",
"evidence": ["docs.crewai.com/en/concepts/agents", "github.com/crewAIInc/crewAI README", "daily.dev AI agents guide"],
"tags": ["entry-gate", "framework-selection"]
},
{
"id": "OP-2",
"name": "Design agent role/goal/backstory triplet",
"trigger": "CrewAI selected; modeling each agent",
"action": "For each agent: write seniority-bearing role title; write goal with template vars + verifiable criteria; write backstory (2-4 sentences) with experience, biases, taboos. Set allow_delegation=False, max_iter=8, verbose=True by default",
"output": "agents.yaml or list of Agent() calls",
"evidence": ["docs.crewai.com/en/concepts/agents"],
"tags": ["design", "prompt-engineering"]
},
{
"id": "OP-3",
"name": "Write Tasks with explicit contracts",
"trigger": "Agents designed; defining work units",
"action": "description = WHAT (not how); expected_output = verifiable assertion; context=[prior_task_refs] explicit; optional output_pydantic for structured outputs",
"output": "tasks.yaml or list of Task() calls",
"evidence": ["docs.crewai.com/en/concepts/tasks"],
"tags": ["design", "contracts"]
},
{
"id": "OP-4",
"name": "Select Process",
"trigger": "Assembling Crew",
"action": "Default Process.sequential. Use Process.hierarchical ONLY with custom manager_agent (never bare manager_llm). If routing logic fits in 5 lines of Python, use Flow instead",
"output": "process parameter + (optional) manager_agent",
"evidence": ["docs.crewai.com/en/learn/hierarchical-process", "towardsdatascience.com Manager-Worker fails", "github.com/crewAIInc/crewAI/discussions/1220"],
"tags": ["control-flow", "high-risk"]
},
{
"id": "OP-5",
"name": "Decide on Memory",
"trigger": "Assembling Crew; checking context requirements",
"action": "Intra-kickoff context => use task.context (memory OFF). Cross-kickoff persistence => memory=True. Production HF => Mem0 backend. Debug period => always OFF",
"output": "memory parameter setting",
"evidence": ["docs.crewai.com/en/concepts/memory", "mem0.ai/blog/crewai-memory-production-setup-with-mem0"],
"tags": ["state-management"]
},
{
"id": "OP-6",
"name": "Install observability and guardrails before production",
"trigger": "Pre-deployment checklist",
"action": "Attach mlflow.crewai.autolog() OR Maxim/Langfuse/Datadog. Set max_rpm, per-agent max_iter, outer kickoff timeout. Pass step_callback for action logs",
"output": "Observable, bounded Crew",
"evidence": ["docs.crewai.com/en/observability/overview"],
"tags": ["production-readiness", "observability"]
},
{
"id": "OP-7",
"name": "Upgrade Crew to Flow when needed",
"trigger": "Conditional routing required OR multiple Crews must chain OR hierarchical proven uncontrollable",
"action": "Subclass Flow[StateModel]. Use @start() entry; @listen(prior) for branches. Each step can be plain Python, single LLM call, or crew.kickoff()",
"output": "Flow class with deterministic spine + autonomous Crews per node",
"evidence": ["docs.crewai.com/en/concepts/flows", "community.crewai.com/t/5710"],
"tags": ["control-flow", "scaling"]
},
{
"id": "OP-8",
"name": "Scope tools per-role",
"trigger": "Multiple tools available across multiple agents",
"action": "Define tools once (BaseTool/@tool). Bind each agent only to role-relevant tools. Avoid blanket tool sharing",
"output": "Per-agent tools=[...] lists",
"evidence": ["docs.crewai.com/en/concepts/tools", "community.crewai.com/t/tool-best-practice-assign-to-agent-or-task/5919"],
"tags": ["design"]
},
{
"id": "OP-9",
"name": "Default allow_delegation=False on all worker agents",
"trigger": "Agent creation, especially in hierarchical mode",
"action": "Workers always allow_delegation=False; only manager_agent (if any) gets delegation. Ensure at least one agent in chain has it disabled to break cycles",
"output": "Delegation-loop-safe Crew",
"evidence": ["github.com/crewAIInc/crewAI/issues/330", "azguards.com delegation-ping-pong", "inkog.io/glossary/crewai-infinite-loop"],
"tags": ["safety", "high-risk"]
},
{
"id": "OP-10",
"name": "Split agents on persistent failure modes",
"trigger": "Agent fails to converge after 3-5 iterations on real cases",
"action": "Classify failure: style inconsistency => fix backstory; missing checklist items / factual errors / format violations => split into specialist + reviewer with structured rubric",
"output": "Refactored crew with sharper role separation",
"evidence": ["daily.dev AI agents guide", "anti-patterns Medium article"],
"tags": ["iteration", "design"]
}
],
"anti_patterns_summary": [
"AP-1 Agent inflation (>5 agents without strong justification)",
"AP-2 Anemic backstory (template/generic persona)",
"AP-3 Default hierarchical manager_llm in production",
"AP-4 Global allow_delegation=True",
"AP-5 No observability before production"
],
"dilemma_cases_summary": [
"DC-1 Agent looping: fix prompt vs split agent",
"DC-2 Sequential vs Hierarchical: when is manager overhead worth it",
"DC-3 Tools: shared all-agents vs role-scoped",
"DC-4 Memory: default-off vs enable",
"DC-5 allow_delegation: on vs off (ping-pong risk)"
],
"key_quotes": [
{"text": "An agent needs agency, otherwise it's just another script.", "source": "João Moura, softwareengineeringdaily.com 2025"},
{"text": "Role, goal, and backstory are required and shape the agent's behavior.", "source": "docs.crewai.com/en/concepts/agents"},
{"text": "Single-agent is right for approximately 80% of cases.", "source": "daily.dev AI agents guide"},
{"text": "Crews enable autonomous collaboration; Flows offer precise, event-driven control.", "source": "docs.crewai.com/en/concepts/flows"}
]
}
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deployment | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy. | Stars: 119k | Updated | Claude CodeCodex | Verified | Install |
| childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release. | Stars: 118k | Updated | Claude CodeCodex | Verified | Install |