Library · topic
Tools & Actions
An agent is only as reliable as the interface you hand it. These notes cover designing tools the model can use correctly — schemas, error messages, idempotency — and the guardrails that keep tool-wielding agents safe when something goes wrong. If you build one thing carefully, make it your tools.
- Deep diveverified 2026-07-28
Layered guardrails for agent actions
Guardrails should be enforced in code around the model: validate inputs and outputs, authorize every tool call against the user and task, constrain ar…
- Deep diveverified 2026-07-28
Defending against prompt injection
Treat retrieved pages, emails, documents, tool outputs, and user-uploaded files as untrusted data, even when they contain text claiming to be system i…
- Deep diveverified 2026-07-28
Reliable tool use
Tool-using agents work best with narrow tools, explicit schemas, actionable error messages, and observable results. Validate arguments before executio…
- Field noteverified 2026-07-14
What an agent loop is
An agent loop repeatedly reads the current state, chooses an action, uses a tool, observes the result, and corrects its plan. The model is only one co…
- Field noteverified 2026-07-14
Context management and compaction
Manage context as a budget: retain the system policy, current objective, unresolved constraints, recent actions, exact error messages, and identifiers…
- Field noteverified 2026-07-14
Durable execution and checkpointing
Long-running agents should execute as resumable state machines whose durable checkpoint records the current step, validated state, completed side effe…
- Field noteverified 2026-07-14
Model Context Protocol
The Model Context Protocol is an open standard for connecting AI applications to external tools and data. MCP uses a client-server architecture: serve…
- Field noteverified 2026-07-14
Production sales and GTM agents
Sales agents should personalize from consented, relevant data, respect suppression lists and communication rules, and avoid inferring sensitive traits…
- Field noteverified 2026-07-14
Retries, timeouts, and idempotent tools
Give every networked tool a connection timeout, an overall deadline, and a bounded retry policy using exponential backoff with jitter only for transie…
- Field noteverified 2026-07-14
Function and tool schema design
Expose narrow, intent-level tools with unambiguous names and descriptions, explicit required fields, typed enums and bounds, and no overlapping functi…