basellmflow

The internal engine that coordinates an agent's interaction with a large language model, including conversation history, tool calls, results, planning, and transfers between agents.

In plain words
What is it for?
Use it to prepare model requests, process model responses, run tools, send tool results back, repeat the exchange, and handle transfers between agents.
Why use it?
It manages the repeated steps needed when a model requests a tool, receives its result, and continues toward a final answer. Keeping this logic in one flow avoids making the agent class overly complex.

Cursor rule

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add rules/altaidevorg/rules-for-ai/basellmflow
Clone the repo
git clone --depth 1 https://github.com/altaidevorg/rules-for-ai
Per session 0 Nothing until a file matches its globs; then the whole rule loads.
When invoked 4,195 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00000 $0.04195
Opus 5 $0.00000 $0.02098
Sonnet 5 $0.00000 $0.00839
Haiku 4.5 $0.00000 $0.00419

Measured 2d ago against content hash 50a14b8d6131, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

basellmflow scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

examples/google-adk/basellmflow.mdc · 298 lines

How it starts

The opening of the file, as written. The whole thing — 298 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Chapter 8: BaseLlmFlow

In the previous chapter, we saw how BaseTool allows agents to extend their capabilities beyond simple text generation. But how does the agent actually manage the conversation flow when deciding to use a tool, calling it, processing the result, and generating the final response? This chapter introduces BaseLlmFlow, the internal engine within an Agent (BaseAgent / LlmAgent) that orchestrates this complex interaction cycle with the underlying BaseLlm.

Motivation and Use Case

The logic required to interact effectively with an LLM, especially when tools, planning, or agent transfers are involved, can become quite complex. It often involves:

  1. Preparing the request with appropriate history, instructions, and tool definitions.
  2. Calling the LLM API.
  3. Parsing the response to check for tool calls, code snippets, plan updates, or final answers.
  4. If a tool is called, executing the tool and packaging the result.
  5. Sending the tool result back to the LLM.
  6. Repeating the process until a final answer is generated.
  7. Handling potential agent transfers.

Hardcoding this intricate logic directly into the LlmAgent class would make it monolithic and difficult to customize or extend. BaseLlmFlow abstracts this control flow, separating the how of LLM interaction from the what defined by the agent's configuration (instructions, tools, etc.). This allows for different interaction patterns (flows) and modular customization through processors.

Central Use Case: An LlmAgent equipped with a search_tool receives the query "What's new with Project IDX?".

  • The LlmAgent internally uses an instance of BaseLlmFlow (likely AutoFlow because it handles tools).
  • The BaseLlmFlow prepares the request, adding the search_tool definition.
  • It calls the BaseLlm.
  • The LLM responds with a request to call search_tool(query="Project IDX news").
  • The BaseLlmFlow parses this, identifies the search_tool Tool (BaseTool), executes it, and gets the search results.
  • It formats the results as a FunctionResponse and adds it to the history.
  • It calls the LLM again with the updated history including the search results.
  • The LLM now generates a summary based on the search results.
  • The BaseLlmFlow receives this final text response and yields the corresponding Event. The BaseLlmFlow managed the entire multi-step interaction loop with the LLM and the tool.

Read the full file on GitHub · 298 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2d ago First seen · 298 lines · 0 tokens per session scan A 50a14b8d6131

Subscribe to this mod's changes

basellmflow is a cursor rule published in the GitHub repository altaidevorg/rules-for-ai (2 stars, last pushed 1y ago), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 4,195 tokens. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.