Call for Papers

We invite research on how language agents find, select, and use resources such as tools, reusable skills and workflows, memories, models, and other agents. We refer to these resources as capabilities. The workshop spans representation and indexing, retrieval and ranking, selection and composition, and evaluation through use.

We welcome research papers on retrieval methods, models, architectures, and systems, as well as position papers, benchmarks and datasets, reproducibility studies, demonstrations, and diagnostic or negative results. This list is not exhaustive.

We welcome work on retrieval augmented generation, memory, tool use, deep research, and systems with multiple agents when capability retrieval, routing, or composition is central to the contribution.

See important dates on the homepage for submission deadlines and workshop timing.

Topics of interest

Representing and indexing capabilities. Methods for representing and indexing tools, APIs, skills, workflows, memories, models, and agents using natural language, schemas, code, demonstrations, execution traces, graphs, ontologies, or embeddings. Topics include learning representations that connect user requests with resource functionality, description and metadata enrichment using usage examples and observed behavior to improve retrieval, and updating indexes as resources change.
Retrieval, routing, and composition. Methods for retrieving and ranking individual resources or compatible combinations for a task, including sparse, dense, hybrid, and generative retrieval, reranking, and joint retrieval. Topics include dependencies between resources and selection that accounts for permissions, reliability, and cost, alongside tradeoffs between retrieval quality, latency, and context budgets.
Retrieval during agent execution. Adapting retrieval to an agent’s state, plan, observations, and previous tool calls. Topics include query reformulation, clarification, task decomposition, retrieval from interaction histories, and learning from execution feedback to improve subsequent retrieval decisions. This includes how retrieved resources are incorporated, invoked, and combined, and how delegation, replanning, recovery, and abstention inform further retrieval.
Evaluating capability use. Benchmarks and methods that distinguish errors in retrieval, incorporation, invocation, composition, and execution. Topics include whether retrieved sets contain the required resources, respect dependencies, and support reliable task completion within resource constraints, as well as confidence calibration. We also welcome studies of generalization to unfamiliar resources, retrieval for tools and skills with little usage history, and robustness across languages, domains, and noisy candidate collections.
Capability registries and deployment. Retrieval in dynamic, federated, and access controlled registries, including Model Context Protocol ecosystems. Topics include provenance, permissions, freshness, monitoring, and caching, with an emphasis on maintaining retrieval quality as resources change or become unavailable. This includes ranking under changing reliability and handling misleading or malicious descriptions that affect retrieval and resource selection.

Work that assumes a fixed, manually selected capability set is outside the workshop’s scope.

Submission guidelines

Draft author guidelines. Paper limits, eligible ARR cycles, and publication arrangements will be confirmed in the final call.

Paper preparation

Long papers: up to 8 pages of main content for substantial research contributions, systems, datasets, or evaluation studies.
Short papers: up to 4 pages of main content for focused contributions, position papers, work in progress, or carefully supported negative results.

References and appendices are excluded from these proposed limits. Prepare a single PDF using the official ACL style files in review mode. Remove author names and affiliations, and anonymize supplementary materials and linked resources.

Clearly describe the work’s limitations, evaluation methodology, and relevant ethical considerations. The main paper should contain the information needed to assess the contribution.

Submission and review

Direct submission: submit through the workshop’s OpenReview portal. Papers will receive double blind review, with three independent reviews planned for each submission.
ARR commitment: papers with completed ACL Rolling Review reviews and a meta-review can be submitted through a separate commitment route. Eligible cycles and commitment instructions will be announced.

The submission portal will be announced here.

Join the program committee

We invite researchers from academia and industry to review for RBD at NAACL 2027. We welcome expertise in information retrieval, natural language processing, and language agents, particularly retrieval of tools, skills, workflows, and memory.

Visit our Program Committee Signup page to share your expertise and preferred reviewing load.

Publication options

Archival: original, unpublished work intended for inclusion in the workshop proceedings, subject to the final publication arrangements.
Nonarchival: work intended for presentation and discussion without inclusion in the proceedings.