Peilivanidis, Vladimiros · Nov 5, 2026
Peilivanidis argues agents should be equipped with modular software capabilities
Markdown · JSON · Original source
In an paper, Vladimiros Peilivanidis argues that agents should use Skillware—modular, versioned software capability bundles—instead of relying on prompts to generate code, guess payloads and remember operational constraints. He presents each bundle as combining a typed JSON input contract, deterministic Python code, task directives, bundled datasets and automated unit tests. Adapters, he says, translate the capabilities into tool declarations for different models, including local systems that run without network . The paper reports an evaluation of 23 frontier and across 368 conditions. Its abstract says small local models equipped with Skillware outperformed unequipped on complex multi-step workflows, while reducing token use and and eliminating parameter errors in those conditions. Peilivanidis uses these results to argue that agent reliability should come from verified, deterministic software rather than larger models or longer prompts. The proposal’s practical significance, as presented, is to move operational rules and execution into testable software bundles rather than leave them in model-generated instructions. The abstract also describes a possible extension from digital systems to IoT, drones and embodied robotics. It does not establish how well the reported results generalize beyond the
Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.
Original source
cs.AI · Article
Source details
Original source date: Nov 5, 2026 (2026-11-05).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-08T18:47:23.050Z. First observed: 2026-11-06T05:07:21.600Z. Retrieved: 2026-11-08T18:46:30.030Z.
Referenced by
Recorded citations and public file attachments. A reference alone does not establish agreement or influence.
No public references recorded here yet.
Limits
This reading is based on the page’s abstract, not the full paper; the abstract does not describe the tasks, evaluation protocol or how results were measured. The reported performance and cost figures are claims made in the abstract; this evidence does not establish how well they generalize beyond the evaluated conditions.
Open question
What tasks and evaluation details would show whether Skillware’s reported results generalize beyond the tested conditions?
Source-stated
Peilivanidis argues that Skillware separates an agent’s intended task from its execution through modular, versioned software capability bundles.
Skillware, an open architectural framework and capability registry that formally separates what an agent wants to do from how that task is executed.
Source-stated
The paper describes bundles combining typed inputs, deterministic code, task directives, datasets and unit tests, with adapters for different models.
each bundle combines a typed JSON input contract, deterministic Python code, task directives, bundled datasets, and automated unit tests.
Source-stated
The abstract reports that equipped small local models outperformed unequipped frontier models on the evaluated multi-step workflows, with lower token use and inference costs and no parameter errors.
small local models (3B to 8B parameters) equipped with Skillware consistently outperform unequipped frontier models on complex multi-step workflows.
Source-stated
Peilivanidis argues that reliability should come from verified, deterministic software rather than larger models or longer prompts.
True agent reliability does not come from larger models or longer prompts, but from equipping logical systems with verified, deterministic software.
Source-stated
The abstract presents extending capability bundles to IoT, drones and embodied robotics as a future trajectory.
from digital logical systems today, to IoT, drones, and embodied robotics tomorrow
Source and contributor context
Peilivanidis, Vladimiros · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
API
An application programming interface: a defined way for software to request data or actions from another system.
Latency
The delay between sending a request and receiving the result.
Open weights
A model whose trained numbers are published so others can download, run and inspect it; the license and training data may still be restricted.
Benchmark
A defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Frontier model
One of the most capable AI models available at a given time; the set shifts as models improve.
Model inference
Using a trained model to produce a prediction or response to new input.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Editorial provenance
Version 1. Publication ID: 2767a2db-464a-4fd2-82ef-7400f21b90c0.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-08T18:47:23.050Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 7b094d00664fd54592160042d60a5a3c6928e8cd3dbbc83909ec134b30fa777b. Source text hash: 8a5818958d9569a49f50e64f5430e25cf77e9ae7aee98d3c71c5b7c90d6ef5af.
Revision and source-check history
Version 1 · 2026-11-08T18:47:23.050Z
Peilivanidis argues agents should be equipped with modular software capabilities
In an paper, Vladimiros Peilivanidis argues that agents should use Skillware—modular, versioned software capability bundles—instead of relying on prompts to generate code, guess payloads and remember operational constraints. He presents each bundle as combining a typed JSON input contract, deterministic Python code, task directives, bundled datasets and automated unit tests. Adapters, he says, translate the capabilities into tool declarations for different models, including local systems that run without network . The paper reports an evaluation of 23 frontier and across 368 conditions. Its abstract says small local models equipped with Skillware outperformed unequipped on complex multi-step workflows, while reducing token use and and eliminating parameter errors in those conditions. Peilivanidis uses these results to argue that agent reliability should come from verified, deterministic software rather than larger models or longer prompts. The proposal’s practical significance, as presented, is to move operational rules and execution into testable software bundles rather than leave them in model-generated instructions. The abstract also describes a possible extension from digital systems to IoT, drones and embodied robotics. It does not establish how well the reported results generalize beyond the
Original source
cs.AI · Article
Source details
Original source date: Nov 5, 2026 (2026-11-05).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-08T18:47:23.050Z. First observed: 2026-11-06T05:07:21.600Z. Retrieved: 2026-11-08T18:46:30.030Z.
Limits
This reading is based on the page’s abstract, not the full paper; the abstract does not describe the tasks, evaluation protocol or how results were measured. The reported performance and cost figures are claims made in the abstract; this evidence does not establish how well they generalize beyond the evaluated conditions.
Open question
What tasks and evaluation details would show whether Skillware’s reported results generalize beyond the tested conditions?
Source-stated
Peilivanidis argues that Skillware separates an agent’s intended task from its execution through modular, versioned software capability bundles.
Skillware, an open architectural framework and capability registry that formally separates what an agent wants to do from how that task is executed.
Source-stated
The paper describes bundles combining typed inputs, deterministic code, task directives, datasets and unit tests, with adapters for different models.
each bundle combines a typed JSON input contract, deterministic Python code, task directives, bundled datasets, and automated unit tests.
Source-stated
The abstract reports that equipped small local models outperformed unequipped frontier models on the evaluated multi-step workflows, with lower token use and inference costs and no parameter errors.
small local models (3B to 8B parameters) equipped with Skillware consistently outperform unequipped frontier models on complex multi-step workflows.
Source-stated
Peilivanidis argues that reliability should come from verified, deterministic software rather than larger models or longer prompts.
True agent reliability does not come from larger models or longer prompts, but from equipping logical systems with verified, deterministic software.
Source-stated
The abstract presents extending capability bundles to IoT, drones and embodied robotics as a future trajectory.
from digital logical systems today, to IoT, drones, and embodied robotics tomorrow
Source and contributor context
Peilivanidis, Vladimiros · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
API
An application programming interface: a defined way for software to request data or actions from another system.
Latency
The delay between sending a request and receiving the result.
Open weights
A model whose trained numbers are published so others can download, run and inspect it; the license and training data may still be restricted.
Benchmark
A defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Frontier model
One of the most capable AI models available at a given time; the set shifts as models improve.
Model inference
Using a trained model to produce a prediction or response to new input.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Editorial provenance
Version 1. Publication ID: 2767a2db-464a-4fd2-82ef-7400f21b90c0.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-08T18:47:23.050Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 7b094d00664fd54592160042d60a5a3c6928e8cd3dbbc83909ec134b30fa777b. Source text hash: 8a5818958d9569a49f50e64f5430e25cf77e9ae7aee98d3c71c5b7c90d6ef5af.
uncertain · 2026-11-08T18:47:23.050Z · anthropic/claude-opus-5.5
Publication: 2767a2db-464a-4fd2-82ef-7400f21b90c0
attribution — pass: The reading correctly attributes the arXiv paper and its arguments to Vladimiros Peilivanidis. The framing keeps the performance and reliability claims as the author's assertions, not established fact.
Authors:Vladimiros Peilivanidis (ARPA Hellenic Logical Systems)
summary — uncertain: The summary's final paragraph breaks off mid-sentence: "It does not establish how well the reported results generalize beyond the". That leaves the limitation incomplete. Smallest repair: finish the sentence, for example "...beyond the evaluated conditions." Central terms such as API payloads, typed JSON input contract, token use, open-weight and frontier models lack plain-language glosses or term-library tooltips, which may block a non-specialist reader. The author's video-game skill-hotkey analogy would help here and could be restored. Title and question are faithful and non-leading.
Much like skill hotkeys in video games that execute deterministically under fixed world rules
date — pass: originalPublished 2026-11-05 matches the arXiv v1 submission date. The later feed entry date was correctly not used.
[Submitted on 5 Nov 2026]
support — pass: Each claim matches its abstract excerpt. Results are scoped to the evaluated conditions and labeled as abstract claims. The extension to IoT, drones and robotics is presented as a trajectory, not a demonstrated result. The limitations correctly note that only the abstract was available. No unsupported benefits or costs are implied.
They cut token usage by up to 84.8%, slash inference costs by up to 97.3%, and eliminate parameter errors entirely.
media — pass: The selected media is the verified direct arXiv abstract page for this paper, which is the most direct access to the source.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivAn application programming interface: a defined way for software to request data or actions from another system.
Definition referenceThe delay between sending a request and receiving the result.
Definition referenceA model whose trained numbers are published so others can download, run and inspect it; the license and training data may still be restricted.
Definition referenceA defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Definition referenceOne of the most capable AI models available at a given time; the set shifts as models improve.
Definition referenceUsing a trained model to produce a prediction or response to new input.
Definition referencearXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivarXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivA defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Definition referenceA defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Definition referencearXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivAn application programming interface: a defined way for software to request data or actions from another system.
Definition referenceThe delay between sending a request and receiving the result.
Definition referenceA model whose trained numbers are published so others can download, run and inspect it; the license and training data may still be restricted.
Definition referenceA defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Definition referenceOne of the most capable AI models available at a given time; the set shifts as models improve.
Definition referenceUsing a trained model to produce a prediction or response to new input.
Definition referencearXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivarXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivA defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Definition referenceA defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Definition reference