procedural-memory

A small case study on poisoning in agent procedural memory shows that a 'proof before reuse' rule—serving a learned skill only after two later sessions reproduce it, showing only the reproduced lines—keeps poisoned skills out of memory (0/30 exposure vs 3/30 for naive memory and 30/30 for replay), with all data, code, and per-session records released.

A continual adaptation framework evolves procedural memory from user traffic for agentic graphic design, improving execution success rates without weight updates or human labels.

Memorable introduces a method to optimize memory in AI agents using embeddings instead of tokens, enabling procedural memory that persists across multiple runs.

This study investigates how procedural memory mismatch in language agents affects web tasks, finding that under controlled conditions, mismatch does not lead to behavioral disruption or errors.

SkillGLoW demonstrates that self-improving agents achieve better performance by storing reusable task-solving procedures rather than memorizing every past task, gaining 17.2 points with a 3.6× more compact library.

A developer shares a prototype memory architecture for multi-agent systems with separate storage for episodic, semantic, and procedural memories and intent-based retrieval, seeking feedback on scalability and design choices.

Google Research's WikiSkill preprint demonstrates that compiling execution history into skill files allows a 9B AI model to outperform a 27B model in benchmarks, highlighting the role of procedural memory and skill transfer in agent performance.

The author questions whether existing agent-memory setups track the success of learned workflows and proposes their approach of using counters and evolution logs, seeking prior art or conventions in the field.

This paper investigates the conditions under which skills enhance LLM agent performance, analyzing success and failure factors through controlled experiments and proposing a taxonomy of skill-use modes.

This paper introduces SkillZip, a contract-preserving graph compression framework that compresses agent skill libraries at the section level, improving retrieval and compression efficiency while preserving procedural contracts and dependency closures.

An analysis of the gap between episodic and procedural memory in LLM agents, citing a new paper (Memp) from Zhejiang University and Alibaba that builds procedural memory from agent trajectories and uses failure signals to revise stored procedures.

A Twitter thread offering strategic advice on building self-learning agents that improve with use, covering dual learning sources, memory types, and data ownership.

The author argues that AI agents lack a mechanism to consolidate scattered operational knowledge into trusted procedural memory, which is a critical blocker for reliable deployment, and asks the community if this is the real bottleneck.

A developer shares their experience building an AI agent with memory using the Anthropic SDK and TypeScript, explaining the differences between working, episodic, semantic, and procedural memory and the challenges of scaling memory for production.

Procedural Memory Distillation (PMD) converts cross-episode signals from reinforcement learning rollouts into reusable procedural memory that is distilled into the policy weights during training, enabling self-improving language models without memory at inference. Experiments show PMD outperforms SDPO by 3.8-5.5% on SCIKNOWEVAL and 7.9-13.6% on LIVECODEBENCH.

This paper introduces Neural Procedural Memory (NPM), a training-free framework that stores procedural skills as activation steering vectors distilled from contrastive historical experience, enabling LLM agents to execute skills without relying solely on textual instructions.

Introduces AFTER, a benchmark of 382 enterprise tasks to evaluate procedural memory in LLM agents, showing that skill transfer improves performance across tasks, roles, and model backbones, with some skills generalizing broadly while others specialize.