---
slug: "ToolInvoke"
name: "ToolInvoke"
type: "Primitives"
layer: "Infrastructure"
ref: "ToolInvoke#1096"
sema_id: "sema:ToolInvoke#mh:SHA-256:1096039316801cc8ef6e2855a6300574bb8596bb8e57c2764fd3bee741fa217e"
tags: ["Infrastructure","Primitives"]
updated: 2026-11-11
mindscape: "bootstrap"
index_url: https://mindscape.world/scapes/bootstrap.json
url: https://mindscape.world/bootstrap/ToolInvoke/
related: ["InputGuard","Context","Sandbox","Task","PolymorphicSolver","PromptChain","ReAct","ToolDiscovery","AgentSandbox"]
---

# ToolInvoke

Execute external tool and observe result

Layer: Infrastructure · Category: Primitives

## Mechanism

The actor emits a structured tool call (function name + arguments), validated by [InputGuard](https://mindscape.world/scapes/bootstrap/InputGuard.md), execution is delegated to an external runtime, and the observation (result or error) is fed back into the actor's [Context](https://mindscape.world/scapes/bootstrap/Context.md). This is the atomic unit of agent-environment interaction. The pattern enables capabilities beyond text generation: code execution, API calls, file operations, web search.

## Invariants

- [Context](https://mindscape.world/scapes/bootstrap/Context.md) Inheritance: Tool Execution [Context](https://mindscape.world/scapes/bootstrap/Context.md) permissions MUST be <= [Task](https://mindscape.world/scapes/bootstrap/Task.md) Constraints. Elevation forbidden.
- Observation Integration: Result must be incorporated into subsequent reasoning
- Sandboxing: Tool execution must respect the capability boundary of its [Sandbox](https://mindscape.world/scapes/bootstrap/Sandbox.md) — the boundary is what makes the permission ceiling enforceable rather than advisory
- Schema Conformance: Tool call must match declared function signature

## Failure modes

- Hallucinated Tools: Actor invokes a tool that doesn't exist.
- Argument Mismatch: Tool call has wrong parameter types or missing required fields.
- Observation Blindness: Actor ignores or misinterprets tool output.

## Caution

Tool results are untrusted input — text inside an observation can carry instructions that steer the agent; treat it as data, never as commands.

## Builds on

- [InputGuard](https://mindscape.world/scapes/bootstrap/InputGuard.md) (Infrastructure; Composes with)
- [Context](https://mindscape.world/scapes/bootstrap/Context.md) (Infrastructure; References)
- [Sandbox](https://mindscape.world/scapes/bootstrap/Sandbox.md) (Infrastructure; References)
- [Task](https://mindscape.world/scapes/bootstrap/Task.md) (Infrastructure; References)

## Used by

- [PolymorphicSolver](https://mindscape.world/scapes/bootstrap/PolymorphicSolver.md) (Mind; Composes with)
- [PromptChain](https://mindscape.world/scapes/bootstrap/PromptChain.md) (Society; Composes with)
- [ReAct](https://mindscape.world/scapes/bootstrap/ReAct.md) (Mind; Composes with)
- [ToolDiscovery](https://mindscape.world/scapes/bootstrap/ToolDiscovery.md) (Society; Composes with)

## Identity

- Sema ID: `sema:ToolInvoke#mh:SHA-256:1096039316801cc8ef6e2855a6300574bb8596bb8e57c2764fd3bee741fa217e`
- Ref: `ToolInvoke#1096`
- Supersedes:
  - `sema:ToolInvoke#mh:SHA-256:643c8651a44f1d33cb98580b8b456c5ec1cdb3b94f3d6ae30e3465fc0e6d45b2`
  - `sema:ToolInvoke#mh:SHA-256:cf0ae58828cde9f495b09af3e4519c3f05162066991d34d5f2505d13c32ba416`
  - `sema:ToolInvoke#mh:SHA-256:011f50770f2599f5017df648ad5920ddb732f7681a9a3b7576cdcf3798ce6baa`
  - `sema:ToolInvoke#mh:SHA-256:aafcc41012592030d7d1e83e91115fbaff1d238b5fc88827efda1649c7a7b502`
  - `sema:ToolInvoke#mh:SHA-256:75cc12a769885fb63993dbb0ffbf0b0878c7e780fe89d6a0d49bd0bc550e60b5`

## Links

- [Pattern page](https://mindscape.world/bootstrap/ToolInvoke/)
- [View in the interactive graph](https://mindscape.world/bootstrap/#ToolInvoke)
- [JSON](https://mindscape.world/scapes/bootstrap/ToolInvoke.json)
- [Index of all patterns](https://mindscape.world/scapes/bootstrap.json)