​

Agentic Patterns (v1.1) — Where TealTiger Fits

Last updated: 2026-03-03

​

What TealTiger v1.1 can do (scope reminder)

TealTiger v1.1 is applied at execution boundaries:

• Model invocation boundary (before/after calling the LLM)
• Tool invocation boundary (before executing a tool/API/function)

Typical outcomes in v1.1-style deployments:

• Allow an action
• Block an action
• Route to human review/approval (if your app implements a review path)
• Redact/transform output (only if your v1.1 integration implements it)

---

​

Pattern 1 — Single-Step Tool Agent

​

What it is

One user request leads to one model call and (optionally) one tool call.

​

Primary risks

• Unsafe tool call (wrong tool / wrong parameters)
• Leaking sensitive data in responses

​

Where to apply TealTiger v1.1

• Tool boundary: validate/allow/deny the tool invocation
• Model boundary: validate response before returning to user

​

Diagram (v1.1 placement)

---

​

Pattern 2 — Planner → Executor (Two-Phase Agent)

​

What it is

The agent creates a plan (steps) and executes step-by-step.

​

Primary risks

• Plan drift (steps exceed intended scope)
• Hard-to-stop mid-flight unless you gate every step

​

Where to apply TealTiger v1.1

• Apply TealTiger at every tool call, not just at the planning phase.

​

Diagram (v1.1 placement)

---

​

Pattern 3 — ReAct / Looping Agent

​

What it is

The agent loops: think → act → observe → think → act …

​

Primary risks

• Infinite loops
• Tool-call storms (cost + blast radius)

​

Where to apply TealTiger v1.1

• Apply TealTiger at every tool boundary.
• Add deterministic loop limits in your app logic (e.g., max steps / max tool calls).

​

Diagram (v1.1 placement)

---

​

Pattern 4 — Retrieval-Augmented Agent (RAG)

​

What it is

The agent retrieves documents/context and then calls the model.

​

Primary risks

• Prompt injection via retrieved content
• Data exposure from unapproved sources

​

Where to apply TealTiger v1.1

If v1.1 does not have a dedicated retrieval boundary, treat retrieval as a tool:

• Govern the retrieval call at the tool boundary
• Govern the final answer at the model boundary

​

Diagram (v1.1 placement)

---

​

Pattern 5 — Autonomous Background Agent

​

What it is

The agent runs without a direct user request (scheduler/event-driven).

​

Primary risks

• No human oversight
• Silent destructive actions

​

Where to apply TealTiger v1.1

• Always gate destructive tools at the tool boundary.
• Prefer “review required” defaults for high-impact tools.

​

Diagram (v1.1 placement)

---

​

Pattern 6 — Human-in-the-Loop Agent

​

What it is

The agent pauses for approval before executing a high-impact action.

​

Primary risks

• Approval fatigue
• Inconsistent review handling

​

Where to apply TealTiger v1.1

• Use TealTiger decisioning to route actions to your existing review flow.

​

Diagram (v1.1 placement)

---

​

Pattern 7 — Multi-Agent Collaboration (v1.1-safe)

​

What it is

Multiple agents coordinate, exchange intermediate results, and divide tasks.

​

Primary risks

• Responsibility diffusion (harder to attribute “why”)
• Cross-agent trust assumptions (one agent passes unsafe instructions to another)
• Increased blast radius (parallel tool calls)

​

Where to apply TealTiger v1.1

• Apply TealTiger per agent, at each agent’s tool boundary.
• Use a shared correlation_id/run_id in your app to keep logs traceable across agents.

​

Diagram (v1.1 placement)

---

​

Summary (v1.1)

• TealTiger v1.1 fits wherever models and tools are invoked.
• For multi-step/looping agents, the rule is simple: gate every tool call.
• For multi-agent systems, gate tool calls per agent and keep traceability via shared correlation IDs.

---

​

Implementation checklist (v1.1)

• Identify your model call site(s) and wrap them
• Identify your tool execution site(s) and wrap them
• Add deterministic loop limits (max steps/tool calls) in your agent runner
• Ensure cross-agent workflows share correlation_id/run_id for traceability
• Log every decision (action + reason code) for audit/debug