Procedural Graphs: Self-Evolving Execution Structures for LLM Agents

arXiv preprint 2609.09153 introduces a Procedural Graph that organizes procedural knowledge into (procedure, relation, procedure) triplets and, at each decision step, localizes an agent's active node and uses a guidance model to translate the surrounding subgraph into step-level situational guidance. The graph is self-evolving: an LLM refiner contrasts failed and successful trajectories, edits topology and attributes, commits edits that preserve or improve held-out validation performance, and retains rejected edits to discourage repetition. The abstract reports gains over memory-based baselines across datasets, task types, and LLMs, and that self-evolution further improves performance without manual engineering.

September 10, 20261 source55% confidence

60Strategic

Strategic score

Evidence strength30

Technical novelty75

Commercial potential55

The new element is a Procedural Graph: procedural knowledge is represented as (procedure, relation, procedure) triplets, paralleling knowledge graphs for what-is questions. At each decision step, the framework localizes the agent's active node, and a guidance model translates the surrounding subgraph into step-level situational guidance that biases the solver's next action without dictating it.¹

Context

The constraint being addressed is that most agents select actions through unconstrained generation over an accumulating history, leaving implicit what to do, in what order, and under which conditions. As trajectories lengthen, this can cause loss of objectives, out-of-order tool invocation, and repeated unproductive actions.¹

Assessment

The claimed mechanism combines two loops: a guidance loop that localizes the active node and injects subgraph-derived step-level guidance, and a self-evolution loop in which an LLM refiner contrasts failed trajectories with successful ones, edits graph topology and attributes, commits edits that preserve or improve held-out validation performance, and retains rejected edits to discourage repetition. Starting from a minimal skeleton, the abstract reports that the loop can match or surpass hand-designed graphs and repair a flawed expert prior.¹

Caveats

What would change the assessment: independent, peer-reviewed replication with numerical results; explicit dataset, task, and LLM specifications; ablations isolating guidance from self-evolution; and evidence that retained rejected edits improve behavior without over-constraining valid action sequences. The abstract reports consistent gains over memory-based baselines and further improvement from self-evolution without manual engineering, but supplies no effect sizes or protocols.¹

Source text

Hugging Face Daily Papers — Procedural Graphs: Self-Evolving Execution Structures for LLM Agents · September 8, 2026

Computer Science > Artificial Intelligence

Title:Procedural Graphs: Self-Evolving Execution Structures for LLM Agents

Abstract:Large language models are increasingly deployed as agents that plan over long horizons and act through external tools. Most agents select actions through unconstrained generation over an accumulating history, leaving implicit the procedural knowledge of what to do, in what order, and under which conditions. As trajectories lengthen, agents can lose track of their objectives, invoke tools out of order, and repeat unproductive actions. We introduce the Procedural Graph: just as a knowledge graph organizes factual knowledge into (entity, relation, entity) triplets for what-is questions, a Procedural Graph organizes procedural knowledge into (procedure, relation, procedure) triplets for what-to-do questions. At each decision step, the framework localizes the agent's active node, and a guidance model translates the surrounding subgraph into step-level situational guidance that biases the solver's next action without dictating it. The graph is self-evolving: an LLM refiner contrasts failed trajectories with successful ones and edits the graph's topology and attributes, committing edits that preserve or improve held-out validation performance while retaining rejected ones to discourage repetition. Starting from a minimal skeleton, the loop builds graphs that match or surpass hand-designed ones. It can also repair a flawed expert prior. Across multiple datasets, task types, and LLMs, the Procedural Graph delivers consistent gains over memory-based baselines, and self-evolution further improves performance without manual engineering.

Current browse context:

Bookmark

Bibliographic and Citation Tools

Code, Data and Media Associated with this Article

Demos

Recommenders and Search Tools

arXivLabs: experimental projects with community collaborators

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.

The term 'self-evolving' and the clause 'retains rejected ones to discourage repetition' assume that failed edit proposals encode generalizable negative constraints. The abstract does not specify how rejected edits are represented, when they expire, or whether retaining them can block later beneficial edits and inflate graph complexity.

'Preserve or improve held-out validation performance' is the acceptance test for graph edits, but the basis of validation is not stated: the metric, split, task distribution, and acceptance threshold are all unspecified. Without that basis, the self-evolution claim could rest on a narrow or noisy validation signal.

'Consistent gains over memory-based baselines' is not accompanied by a definition of those baselines or by effect sizes, significance tests, or evidence that gains persist across task types. The abstract's comparative claim therefore has limited evidential weight on its own.

The source text above remains the property of its publisher and is reproduced for commentary.

Hugging Face Daily Papers — Procedural Graphs: Self-Evolving Execution Structures for LLM Agents

September 8, 2026endistinct angle

Procedural Graphs: Self-Evolving Execution Structures for LLM Agents

↑ Back to top

Procedural Graphs: Self-Evolving Execution Structures for LLM Agents

September 10, 20261 source55% confidence

60Strategic

Strategic score

Evidence strength30

Technical novelty75

Commercial potential55

Context

Assessment

Caveats

Source text

Hugging Face Daily Papers — Procedural Graphs: Self-Evolving Execution Structures for LLM Agents · September 8, 2026

Computer Science > Artificial Intelligence

Title:Procedural Graphs: Self-Evolving Execution Structures for LLM Agents

Abstract:Large language models are increasingly deployed as agents that plan over long horizons and act through external tools. Most agents select actions through unconstrained generation over an accumulating history, leaving implicit the procedural knowledge of what to do, in what order, and under which conditions. As trajectories lengthen, agents can lose track of their objectives, invoke tools out of order, and repeat unproductive actions. We introduce the Procedural Graph: just as a knowledge graph organizes factual knowledge into (entity, relation, entity) triplets for what-is questions, a Procedural Graph organizes procedural knowledge into (procedure, relation, procedure) triplets for what-to-do questions. At each decision step, the framework localizes the agent's active node, and a guidance model translates the surrounding subgraph into step-level situational guidance that biases the solver's next action without dictating it. The graph is self-evolving: an LLM refiner contrasts failed trajectories with successful ones and edits the graph's topology and attributes, committing edits that preserve or improve held-out validation performance while retaining rejected ones to discourage repetition. Starting from a minimal skeleton, the loop builds graphs that match or surpass hand-designed ones. It can also repair a flawed expert prior. Across multiple datasets, task types, and LLMs, the Procedural Graph delivers consistent gains over memory-based baselines, and self-evolution further improves performance without manual engineering.

Current browse context:

Bookmark

Bibliographic and Citation Tools

Code, Data and Media Associated with this Article

Demos

Recommenders and Search Tools

arXivLabs: experimental projects with community collaborators

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.

The source text above remains the property of its publisher and is reproduced for commentary.

↑ Back to top