Skip to content

agentsope/SkillAlchemy / skills/agentsop-tool-scoping

agentsop-tool-scoping

Verified

Enhancement overlay for multi-agent / tool-using coder agents. Encodes the per-agent tool- scoping discipline that role-based frameworks (CrewAI, LangChain) document only as a passing best-practice: which agent gets which tool, and why blanket-sharing every tool to every agent is a correctness and blast-radius risk. Activates when an agent system has tools AND there is more than one agent (or one agent holding many tools). Treat a tool as a capability grant; scope by least-privilege. ENHANCE overlay — read alongside [[crewai]], [[agentsop-http-tool-wrapping]], [[agentsop-llm-tool-idempotency]]. Search keywords: which tools per agent, least-privilege agent, agent tool access, tool permissions, limit agent tools, scope tools to roles.overlay_type: enhancement

FileKindSize
README.mdmarkdown2.5 KB
SKILL.mdskill md18.0 KB
intermediate/operation_candidates.jsonconfig4.4 KB
references/R1-source-evidence.mdmarkdown4.8 KB

intermediate/operation_candidates.json

View on GitHub
{
  "skill": "tool-scoping",
  "version": "0.1.0",
  "overlay_type": "enhancement",
  "enhances": ["crewai", "langchain", "langgraph"],
  "task_id": "T1",
  "task_title": "Tool definition + per-agent binding (scoping discipline)",
  "core_mental_model": "A tool is a capability grant; scope by least-privilege. Tool definition (reuse) and tool binding (per-agent, minimal) are separate decisions.",
  "operation_candidates": [
    {
      "id": "OP-1",
      "name": "Least-privilege role->tool mapping",
      "trigger": "Assembling a multi-agent crew/graph from a shared tool pool",
      "action": "Start each agent at tools=[]; add a tool only when the role requires that agent to exercise it; reuse definition, bind per-role",
      "output": "Minimal per-agent tools list + binding matrix",
      "evidence": ["crewai-sop:DC-3", "langgraph-sop:Step4"],
      "selected": true
    },
    {
      "id": "OP-2",
      "name": "Side-effect tool guarding",
      "trigger": "A tool performs a write or destructive operation",
      "action": "Bind to one agent only; add HITL/approval guard before the effect; make op idempotent",
      "output": "Single-owner, guarded, idempotent side-effectful tool",
      "evidence": ["langgraph-sop:Step5", "langgraph-sop:Case4"],
      "crosslinks": ["llm-tool-idempotency", "http-tool-wrapping"],
      "selected": true
    },
    {
      "id": "OP-3",
      "name": "Shared-registry vs per-agent binding decision",
      "trigger": "A tool registry plus N agents; must decide wiring",
      "action": "Share at definition layer; decide binding per-agent via OP-1; never registry.all() on every agent",
      "output": "One definition source, N minimal per-agent bindings",
      "evidence": ["crewai-sop:DC-3"],
      "selected": true
    },
    {
      "id": "OP-4",
      "name": "Per-agent tool-count limit",
      "trigger": "Agent tool list grows >=8 or starts mis-selecting tools",
      "action": "Cap ~8 tools/agent; split role, add routing sub-agent, or demote read-only tools to context",
      "output": "Every agent <= ceiling; better selection accuracy",
      "evidence": ["crewai-sop:6.1-AP-1"],
      "rationale": "too many tools degrades selection by same mechanism as too many agents degrades coordination",
      "selected": true
    },
    {
      "id": "OP-5",
      "name": "Zero-tool synthesis agents",
      "trigger": "Agent only consumes upstream output (writer/reporter/judge)",
      "action": "Bind no tools; feed via task context / state",
      "output": "Tool-free agent that cannot wander into capabilities",
      "evidence": ["crewai-sop:DC-3"],
      "selected": true
    },
    {
      "id": "OP-6",
      "name": "Binding-matrix audit",
      "trigger": "Pre-ship or security review",
      "action": "Build agents x tools matrix; flag write/destructive bound to >1 agent or unguarded; flag agents over count limit",
      "output": "Reviewable capability matrix + remediation list",
      "evidence": ["SKILL:Phase4"],
      "selected": true
    },
    {
      "id": "OP-7",
      "name": "Tighten scope as fastest misuse fix",
      "trigger": "Agent observed using a tool outside its role",
      "action": "Remove the tool from that agent's binding rather than prompt-engineering 'do not use X'",
      "output": "Structurally prevented misuse",
      "evidence": ["crewai-sop:DC-3"],
      "selected": true
    }
  ],
  "dilemma_cases": [
    {
      "id": "DC-1",
      "title": "Shared tool registry vs per-agent tool sets",
      "source": "crewai-sop:DC-3"
    },
    {
      "id": "DC-2",
      "title": "The 20-tool agent that picks the wrong tool",
      "source": "crewai-sop:6.1-AP-1 + langgraph-sop:Step4/Case4"
    },
    {
      "id": "DC-3",
      "title": "Should a destructive tool ever be shared across agents?",
      "source": "langgraph-sop:Step5/Case4 + crewai-sop:DC-3"
    }
  ],
  "anti_patterns": [
    "blanket tool sharing (tools=[all] on every agent)",
    "no guard on destructive tools",
    "one mega-agent with 20 tools",
    "prompt-patching tool misuse instead of removing the binding",
    "sharing the destructive tool itself instead of its definition",
    "binding tools to synthesis agents"
  ],
  "cross_framework": {
    "crewai": "Agent(tools=[...]) per agent",
    "langgraph": "model.bind_tools([...]) per node",
    "openai_assistants": "per-Assistant tools array",
    "claude_tool_use": "per-request tools list"
  },
  "crosslinks": ["crewai", "http-tool-wrapping", "llm-tool-idempotency"]
}
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 471kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall