Search arXiv⌕ Search

arXiv subjects

Haohan Wang

Publications and source records attributed to Haohan Wang.

At least 19 recordsLinked to original sources
arXiv2026

Characterizing Adversarial GEO via Implicit Brand Injection

Generative search engines increasingly shape how users discover products, services, and commercial entities by generating answers from retrieved web content. This creates an opportunity to influence which brands appear in these answers by modifying the content available to the retrieval system. In this work, we study an adversarial generative engine optimization (GEO) setting in which a predefined brand is added to an article through a single contextually integrated mention. The challenge is to make the mention fit the surrounding article without making the content overtly promotional, while keeping it relevant enough to be retrieved and to influence the generated answer. To study this, we introduce Implicit Brand Injection (IBI), a multi-agent framework that identifies insertion locations, rewrites the selected sentence, verifies the edit, and selects the final candidate. To evaluate IBI at scale, we construct IBIBench, which contains 1,000 web articles and 60 synthetic brands across 12 diverse domains, with both endemic and non-endemic brand placements. We compare IBI with Direct Prompting and Best-of-N with ranking under an equivalent budget, evaluating insertion quality, target-chunk retrieval, and downstream brand propagation. Compared with the baselines, IBI produces insertions that better preserve semantics and disrupt perplexity less, while improving Hit@k across all evaluated cutoffs, query regimes, and placement settings. IBI achieves up to 82.2% Hit@10 and 0.74 MRR@10, while Brand Mention Rate (BMR) and Conditional Propagation (CP) reach up to 62.7% and 75.9%, respectively. These results show that a single brand mention can increase both retrieval exposure and brand visibility in generated responses, highlighting the need for defenses against implicit content manipulation in generative search.

cs.IR↗
arXiv2026

Can Test-Time Reconstruction Close the Loop on Latent Reasoning?

Recent work moves intermediate reasoning from natural-language traces into latent or cache-level representations to reduce token overhead and avoid a discrete communication bottleneck. However, this shift also removes a key advantage of textual reasoning: intermediate states are no longer inspectable, making it difficult to determine whether a latent state still preserves the constraints of the original query. As a result, latent reasoning typically operates in an open loop, where a latent state is produced and consumed without an input-anchored fidelity check. We propose ReLAT (Reconstruction-Guided Latent Reasoning At Test Time), a self-supervised test-time training method that closes this loop using the query itself as the reference. Our key observation is that if a latent state faithfully represents a query, the query should be recoverable from it; if the query cannot be recovered, the latent state has lost task-relevant information. ReLAT operationalizes this principle by constructing a differentiable Question -> Latent Thought -> Question cycle and optimizing query reconstruction loss through the latent thought before answer generation. This anchors opaque latent computation to the problem specification it is supposed to represent. Across mathematical reasoning, knowledge QA, and code generation benchmarks on the Qwen family, ReLAT consistently improves over single-model inference, text-based collaboration, open-loop latent collaboration, and alternative test-time training objectives. On Qwen3-8B, ReLAT raises AIME 2024 accuracy from 56.7% to 73.3%, a 16.6-point gain over the strongest open-loop latent baseline.

cs.AI↗
arXiv2026

Can the Library Speak? Self-Advertised Method Selection for Formal Proving

LLM-based formal provers can retrieve relevant lemmas and prior proofs, but relevance alone does not say whether a mathematical method can be used on the current theorem. A method has prerequisites, a target, an intended action, and obligations that its use leaves to prove. Methods that look equally related to a theorem may therefore differ substantially in whether they offer a plausible next step. We formulate this as an applicability-aware method-selection problem and introduce self-advertisement: before candidates are ranked, a model generates a problem-specific proposal for each one, stating what part of the goal it targets, what action it would take, and what conditions that action requires. We organize 82 reusable methods from Putnam 2000-2014 as Method Contracts, which pair applicability descriptions with Mathlib anchors, a checked example or scaffold, and expected proof obligations. A single batched call elicits proposals across the library; vague or unsupported proposals are demoted, yielding a ranked shortlist accompanied by inspectable claims about each candidate's use. We analyze when similarity-based representations cannot distinguish methods with different applicability, how errors in applicability estimates affect shortlist quality, and what a checked scaffold guarantees under its stated assumptions. Against lexical, embedding, and embedding-plus-LLM reranking baselines, self-advertisement achieves 95.0% hit@5 on Putnam 2015-2025, compared with 84.2% for the strongest reranker. On IMO ProofBench, it achieves 91.7% compared with 88.3%. These results indicate improved coverage of annotated methods in the retrieved shortlists, particularly on Putnam.

cs.AI↗
arXiv2026

Do GEO Gains Survive the Live Web? Evaluating In-the-Wild Visibility Gains of Generative Engine Optimization via LIVE-GEO

Generative engine optimization (GEO) rewrites web content so that LLM-based search systems, also termed generative engines (GEs), rank it higher in their answers and recommendations. However, existing GEO evaluations typically rely on directly supplied text or controlled retrieval pipelines, and whether GEO gains survive the live Web remains unclear, leaving a gap in understanding how GEO strategies affect the visibility of live webpages when accessed and ranked by GEs. To address this gap, we introduce LIVE-GEO, a benchmark for Live, In-the-wild Visibility Evaluation for GEO. LIVE-GEO publishes GEO-rewritten products as live webpages. We use LIVE-GEO to evaluate five existing GEO strategies (CORE, C-SEO, E-GEO, AgenticGEO, and Statistics Addition), applying them to the product descriptions and reviews of 80 fictional brands across 18 product families. We run experiments on three GEs (ChatGPT with GPT-6, Gemini 3.6 Thinking, and Grok in Expert mode), four direct-text API evaluators (GPT-4o, DeepSeek, Kimi, and Claude), and a direct-text local evaluator (Qwen2.5-3B-Instruct). With frozen $\pm0.2\%$ prices, most direct-text evaluators and GEs receiving product URLs give CORE the highest Top-1 rate, reaching 66.3%-76.3% on the GEs, yet other strategies vary in relative performance, and the GEs select the same top product for only 48.8%-66.3% of brands. We also find that ChatGPT is sensitive to the order of candidate URLs: CORE's Top-1 rate is 73.8% and 93.8% under two opposite fixed orders, and the randomized order used by LIVE-GEO yields 82.5%, between the two. In the numerical-faithfulness check, the JEV verifier agrees with the reference labels on 90.7% of 194 numerical benchmark cases and with human judgments on all 50 sampled descriptions, and it flags 86 of the 400 rewritten descriptions, including all 80 Statistics Addition descriptions.

cs.IR↗
arXiv2026

SCI-Defense: Detecting Persuasion-Structured Manipulation of LLM-Based Ranking in Generative Engines

Generative engine optimization (GEO) has reshaped how sellers compete for visibility in LLM-generated rankings. By embedding persuasive arguments in the content they control, sellers can influence how generative engines rank products and web pages and push their own items to the top. Existing defenses either detect harmful intent or unnatural language, which fluent GEO content often lacks, or rewrite all candidates, which alters legitimate listings without identifying the manipulated ones. However, these approaches overlook the persuasive structure through which GEO attacks influence LLM-based rankings. To address this gap, we present SCI-Defense, a framework for detecting persuasion-structured ranking manipulation, in which an LLM judge scores five persuasion devices grounded in research on influence and deception: authority, social proof, scarcity and urgency, comparative claims, and narrative purposiveness. Two lightweight gates for unnatural strings and inflated relevance complement the judge, and flagged candidates are demoted rather than rewritten, each with a record of the candidate and the criterion that triggered the flag. On 1,152 optimized CORE attacks across six Amazon product categories and six MS MARCO web domains, SCI-Defense detects 89.6% of reasoning attacks and 96.2% of review attacks at a 1.2% false positive rate and blocks 96.1% of the attacks that pushed their target into the top three. On a 120-query subset, paraphrasing and truncation block a similar share of attacks but push 4.0--4.5% of honest items out of the top three, compared with 0.6% for SCI-Defense.

cs.LG↗
arXiv2026

Where Does Rhetorical Distortion Live? A Multi-Platform, Multi-Language Measurement of Influencer Content

Much of the manipulation in influencer content is not false but disproportionate, framing accurate facts as historic, urgent, or catastrophic. Existing approaches either verify factual claims or detect clickbait and propaganda in single-platform, single-language text; however, they overlook where such framing originates, because they treat a post and the replies beneath it as one unit. To address this gap, we measure rhetorical distortion with one instrument applied identically across six platforms and two languages, classifying creator posts and their replies separately. Our three-tier pipeline detects five literature-grounded dimensions (Significance Inflation, Anxiety Manufacturing, Novelty Claims, Loaded Language, and Temporal Distortion): bilingual lexical rules propose candidates, negative filters cancel known false positives, and an LLM verifies or relabels the candidates. This design keeps labels transparent and comparable across platforms, calls the LLM on only 14% of posts, and supports validation against human annotators and a second model. On 13,060 posts from 330 accounts on RSS/newsletters, YouTube, Bluesky, Reddit, Weibo, and X/Twitter, platform is far more strongly associated with account-level distortion than content category (partial $η^2=0.49$ vs. $0.09$), and X/Twitter scores highest (DI 7.74 vs. 0.78 for RSS). On X/Twitter, the top-5 replies under a post carry 3.5x the distortion of the tweet they answer (1.5x at equal length), while creator tweets alone are indistinguishable from YouTube and Bluesky; two human annotators and GPT-4o confirm this gap. Measured distortion on threaded platforms thus lives largely in the audience layer, which calls for measuring creator and reply text separately.

cs.SI↗
arXiv2026

Controlling Output Rankings in LLM-Based Search via Generative Engine Optimization

The way customers search for and choose products is changing with the rise of large language models (LLMs). LLM-based search, or generative engines, provides direct product recommendations to users, rather than traditional online search results that require users to explore options themselves. However, these recommendations are strongly influenced by the initial retrieval order of LLMs, which disadvantages small businesses and independent creators by limiting their visibility. In this work, we propose CORE, an optimization method that \textbf{C}ontrols \textbf{O}utput \textbf{R}ankings in g\textbf{E}nerative Engines for LLM-based search. Since the LLM's interactions with the search engine are black-box, CORE targets the content returned by search engines as the primary means of influencing output rankings. Specifically, CORE optimizes retrieved content by appending strategically designed optimization content to steer the ranking of outputs. We introduce three types of optimization content: string-based, reasoning-based, and review-based, demonstrating their effectiveness in shaping output rankings. To evaluate CORE in realistic settings, we introduce ProductBench, a large-scale benchmark with 15 product categories and 200 products per category, where each product is associated with its top-10 recommendations collected from Amazon's search interface. Extensive experiments on four LLMs with search capabilities (GPT-4o, Gemini-2.5, Claude-4, and Grok-3) demonstrate that CORE achieves an average Promotion Success Rate of \textbf{91.4\% @Top-5}, \textbf{86.6\% @Top-3}, and \textbf{80.3\% @Top-1}, across 15 product categories, outperforming existing ranking manipulation methods while preserving the fluency of optimized content.

cs.CL↗
arXiv2026

GenoMAS: Guided Multi-Agent Programming for Scientific Discovery from Genomic Data

Gene expression analysis holds the key to many biomedical discoveries, yet extracting insights from raw transcriptomic data remains challenging due to the complexity of multiple large, semi-structured files and the need for extensive domain expertise. Current approaches either rely on inflexible workflows that fail in edge cases or fully autonomous agents that lack the precision required for rigorous scientific inquiry. GenoMAS addresses this by presenting a team of LLM-based agents that integrates structured workflows with autonomous adaptability, orchestrating diverse specialized agents through typed message-passing protocols. At its core lies a guided-planning framework where programming agents decompose task guidelines into Action Units and dynamically choose to advance, revise, bypass, or backtrack, maintaining logical coherence while adapting to the idiosyncrasies of genomic data. On the GenoTEX benchmark, GenoMAS achieves 89.13% CSC for data preprocessing and 60.48% $F_1$ for identifying trait-associated genes, surpassing prior art by 10.61% and 16.85%. Beyond transcriptomics, it outperforms strong agent baselines on ScienceAgentBench and DiscoveryBench, two independent bioinformatics benchmarks. Qualitative analysis shows that GenoMAS identifies gene-phenotype associations that are biologically plausible and corroborated by the literature.

cs.AI↗
arXiv2026

Gap Detection: Mitigating Homogenization through Demand and Defensibility

Language models often produce semantically similar answers to open-ended questions, even when many plausible alternatives exist. Generating useful alternatives requires identifying both what the query calls for and what existing responses already cover. We introduce Gap Detection, a framework that searches for relevant but unoccupied directions in a shared text embedding space. Its objective, GapScore, combines query relevance (Demand) with dissimilarity to the nearest existing item (Defensibility). The framework optimizes this product on the unit sphere, maps candidate directions to concept words with a language model, and uses task-specific filtering and constrained prompting to turn those concepts into text. On five curated topics and 90 INFINITY-CHAT queries, the reported mean relative reductions in output Collision are 12.8% and 10.8%, respectively. We also instantiate the framework for generative engine optimization over four product markets and for iterative synthetic-data augmentation. In the latter setting, with 100 training examples per method, Gap Detection reduces Collision from 0.324 to 0.298 relative to undirected accumulation and increases the Vendi Score from 27.16 to 29.20; the accuracy difference is one example on a 40-example test set. Component ablations expose the trade-off between semantic separation and relevance, while also showing that better direction scores do not uniformly yield better downstream metrics. These results support explicit gap search as a practical mechanism for diversifying content, with task validity checked separately from geometric novelty.

cs.CL↗
arXiv2026

Prompt Stability Matters: Evaluating and Optimizing Auto-Generated Prompt in General-Purpose Systems

Automatic prompt generation plays a crucial role in enabling general-purpose multi-agent systems to perform diverse tasks autonomously. Existing methods typically evaluate prompts based on their immediate task performance, overlooking the intrinsic qualities that determine their reliability. This outcome-centric view not only limits interpretability but also fails to account for the inherent stochasticity of large language models (LLMs). In this work, we bring attention to prompt stability-the consistency of model responses across repeated executions-as a key factor for building robust and effective prompt generation systems. To quantify this, we propose semantic stability as a criterion for assessing the response consistency of prompts. Based on the proposed metric, we developed the first stability-aware general-purpose prompt generation system that leverages stability feedback to iteratively enhance both prompt quality and system-level performance. Furthermore, we establish a logical chain between prompt stability and task success by analyzing the structural dependencies within our system, proving stability as a necessary condition for effective system-level execution. Empirical results across general and domain-specific tasks demonstrate that our stability-aware framework improves both accuracy and output consistency. By shifting the focus from one-off results to persistent reliability, our work offers a new perspective on prompt design and contributes practical tools for building more trustworthy general-purpose systems. The UI and source code for our system are publicly available at: https://xucheng63.github.io/MyDataPilot/.

cs.AI↗
arXiv2026

ThreadBench: Benchmarking Realism in Agentic Simulation of Real-world Discussions

LLM agents are increasingly used to simulate real-world interactions, but it remains unclear whether simulated behaviors preserve the content patterns and interaction dynamics of real human behaviors. Existing evaluations often focus on individual comments or task performance, rather than whether simulated behaviors match human behavior at thread level. In this paper, we use Reddit discussions as a concrete testbed for real-world social simulation. Reddit provides public, topic-grounded, multi-party interactions with diverse opinions, experiences, and interaction patterns, making it a useful setting for evaluating whether LLM agents reproduce the distributional and structural properties of real online discussions. We introduce ThreadBench, a benchmark for Reddit discussion simulation built from 9,817 real Reddit threads spanning 8 diverse domains and 116 subdomains. ThreadBench uses statistical analyses to compare generated and real discussions across 12 metrics grouped in 4 families: uniformity, expression, tone, and interaction. We further develop Thread-Prompt-based Discussion Generation (ThreadPDG), a Planner-Writer framework that learns how a real discussion grows. The Planner produces thread- and comment-level plans from a real seed discussion, and the Writer generates comments from these plans. For each seed, we generate multiple candidate threads and combine their best branches using thread-level statistics. Across all domains and multiple LLMs, ThreadPDG achieves ThreadBench pass rates of 58-100%, compared with 0-17% for the baselines. Human evaluation with 436 annotators further shows that ThreadPDG is not distinguished from real discussions in 78.7% of pairs, outperforming existing simulation methods.

cs.MA↗
arXiv2026

Evolving Creatives with Clicks: CTR-Driven Advertising image Editing via Attribute Disentanglement and Preference Alignment

The optimization objective for e-commerce advertising creatives has shifted from aesthetic quality to online business metrics like click-through rate (CTR). Recent methods generate appealing ad images from product images by constructing CTR-based reward signals and refining prompt generation via Reinforcement Learning. Such paradigms equally penalize all attributes within an underperforming image, even though some are actually beneficial to clicks. In this work, we propose Evolving Creatives with Clicks (ECC), which reformulates creative optimization as fine-grained editing rather than generation from scratch. ECC selectively refines attributes that hurt CTR while preserving those that contribute positively. Specifically, faced with a rich attribute space across massive images, ECC first identifies 28 attributes significantly correlated with CTR via ANOVA and Random Forest regression. Based on attribute contrasts between images with disparate CTRs, we instruction-tune a prompt model to generate editing suggestions, then refine it with real click preferences since reference-derived suggestions are not uniformly beneficial. Furthermore, we introduce Consistent Preference Optimization (CPO) as ECC's preference optimization strategy, which enforces consistency across semantically equivalent shuffled suggestions to mitigate DPO overfitting and improve preference generalization. Extensive experiments demonstrate that ECC significantly improves the CTR of ad images, outperforming state-of-the-art models by a substantial margin. The complete data pipeline, including attribute extraction and analysis, the resulting dataset, and source code will be released upon acceptance.

cs.CV↗
arXiv2026

Harness Engineering for Software Engineering via Modular Executable Dev-Primitives

Large language models (LLMs) equipped with terminal access have demonstrated strong capabilities in automating software engineering tasks. However, existing agents remain brittle on long-horizon workflows, where they must repeatedly reconstruct program state scattered across source files, configurations, tests, dependencies, and runtime behavior, leading to increasingly long interaction histories, context explosion, and semantic drift. Large repositories further complicate the identification of task-relevant components. To address these challenges, we introduce \textbf{Dev-Primitives} (\emph{Development Primitives}), a modular and executable abstraction that transforms repository components from passive software artifacts into active participants in software engineering. Each Dev-Primitive pairs a repository artifact with a resident LLM, which gives the artifact an agent-native interface grounded in its own implementation and dependencies, enabling natural-language reasoning, inter-component communication, and localized self-modification. Building on Dev-Primitives, we propose \textbf{HERMES}, a Harness Engineering framework for software engineeRing via Modular Executable Dev-PrimitiveS, which instantiates these primitives at repository scale through a dependency-aware dynamic activation mechanism and a bug diagnosis mechanism that maps execution evidence back to the components that must be revised. Extensive experiments on four software engineering benchmarks demonstrate that HERMES outperforms matched baseline harnesses by 12.4\% on average. Moreover, when paired with strong activation and diagnosis models, HERMES, even with Qwen3-8B Dev-Primitives, remains within 4.5\% of the homogeneous GPT-5.6 Sol configuration across all four benchmarks, while reducing inference cost by 26.2\% on Terminal-Bench 4.0, highlighting the importance of harness design in software engineering agents.

cs.SE↗
arXiv2026

Large-scale Repository Engineering via Agent-Native Reusable Code Primitives

Large language models equipped with development environments have moved code generation toward repository-scale construction, yet building complete repositories remains difficult because interacting modules, interfaces, configurations, tests, and dependencies must work together. We introduce Code Primitives, agent-native reusable executable components with interface contracts, dependency closures, validation tests, and provenance. Each primitive uses a resident LLM to assess relevance and adapt its implementation, interfaces, and dependencies to the target repository, and we organize 1,424 validated primitives in CodeFace, a searchable library for repository construction. We introduce LEGO (Large-scale repository Engineering via aGent-native reusable cOde primitives), which activates task-relevant primitives, integrates their adapted implementations with task-specific code while resolving cross-component constraints, and revises the result against executed tests. To measure construction end to end, we build LEGO-REPO, a benchmark of 522 executable reconstruction tasks spanning seven software domains, 22 capability tracks, and five difficulty levels, scored against native test suites between an empty-package floor and original-source ceiling. The strongest of 13 evaluated backbones reaches a delivery score of 0.318 and scores zero on 41.0% of tasks; LEGO improves all 13 by 0.1474 on average and raises GPT-5.6-terra from 0.3180 to 0.5134 (+61.4%). In controlled comparisons, adapted primitives outperform retrieved code supplied as context or vendored unchanged. The effect persists against independent repository agents, across three external benchmarks, and with a disjointly re-mined CodeFace; GPT-OSS-20B for adaptation and diagnosis retains 95.1% of the homogeneous score at 24.0% lower cost.

cs.SE↗
arXiv2026

TextReg: Mitigating Prompt Distributional Overfitting via Regularized Text-Space Optimization

Large language models (LLMs) are highly sensitive to the prompts used to specify task objectives and behavioral constraints. Many recent prompt optimization methods iteratively rewrite prompts using LLM-generated feedback, but the resulting prompts often become longer, accumulate narrow sample-specific rules, and generalize poorly beyond the training distribution. We study this failure mode as prompt distributional overfitting and argue that it reflects a lack of representation control in discrete text-space optimization. We formalize this view through representational inefficiency, a dual-factor measure that decomposes prompt inefficiency into capacity cost and scope narrowness, attributing distributional prompt overfitting to their coupled growth during optimization. We propose TextReg, a regularization framework that realizes a soft-penalty objective through regularized textual gradients, combining Dual-Evidence Gradient Purification, Semantic Edit Regularization, and Regularization-Guided Prompt Update. Across multiple reasoning benchmarks, TextReg substantially improves out-of-distribution (OOD) generalization, with accuracy gains of up to +11.8% over TextGrad and +16.5% over REVOLVE.

cs.CL↗
arXiv2026

Objects Without Morphisms: What LLMs for Mathematics Do Not Represent

Large language models (LLMs) have reached expert-level performance on competition mathematics largely through the volume of search placed around them: candidate solutions are sampled in quantity and retained only when an external criterion accepts them. Such a procedure improves the outcome that survives it while leaving untouched what the model represents. We examine that question where no external criterion exists: translating statements between the dialects of neighbouring subfields, where fidelity turns on the level of generality at which content is asserted. The source leaves that level implicit in its vocabulary, so a faithful translation must recover it from the relation between the theories. We introduce an instrument that codes truth, content and scope in separate blind queues, with a judge-free measure of whether a rewrite states the hypothesis implicit in its source, and establish its sensitivity with a planted-positive control. Across seven models from four families, translating towards the general framing widens the domain of quantification in 60.6% of rewrites and narrows it in none; translating towards the concrete framing narrows it in 28.3% and widens it in 0.3%. The hypothesis that would prevent it is stated in 21.6% of model rewrites and 4.2% of human statements. Capability does not govern the asymmetry: it appears in every model tested, and the most capable widens least. It replicates on the half of the benchmark held out by a pre-registered rule, and on statements written by mathematicians. Instructing a model to state every hypothesis it requires raises that rate but not its sensitivity to direction. We argue that these systems have acquired an object-level correspondence between subfield vocabularies without the constraint under which a translation between theories carries hypotheses to hypotheses.

cs.LG↗
arXiv2026

From Knowledge Access to Source Learning: Developing Source-Specific Competence

Large language model (LLM) agents increasingly rely on persistent external sources to solve sequences of knowledge-intensive tasks. Existing methods improve how source content is accessed and organized, while agent-memory systems preserve reusable knowledge from prior interactions, but repeated use of the same source is still largely treated as repeated access rather than an opportunity to progressively improve understanding of that source. We study source learning: developing reusable source-specific competence over a persistent authoritative source. We represent this competence with a persistent source model that captures reusable understanding of the source, including how its knowledge is structured, interpreted, and applied. To construct and progressively refine such models, we propose SourceLearn, which combines two complementary learning mechanisms. Self-Directed Source Learning identifies what remains incompletely understood and adaptively revisits the source, while Task-Guided Source Learning uses downstream experience to reveal local representational gaps and recurring needs in how source knowledge should be organized. In both cases, learning signals determine what should be reconsidered, while persistent updates are reconstructed from the authoritative source. Across five benchmarks and three LLM backends, SourceLearn achieves the best performance in 13 of 15 settings, with gains of up to 22.6 points over Hybrid RAG and substantial overall improvements over static source representations and experience-based memory baselines.

cs.CL↗
arXiv2026

Composing Task-specific Agent Harnesses at Test Time with Reusable Primitives

Agent harnesses govern how large language models (LLMs) gather context, invoke tools, verify results, preserve state, and terminate, largely affecting agent performance. However, the value of each harness mechanism can differ across heterogeneous tasks: a mechanism that improves one task may impose overhead or context distraction on another, leading to the suboptimality of a global harness. We characterize this suboptimality as a mismatch induced by fixed mechanism choices, motivating task-specific harness construction. Nonetheless, generating harness code for each task introduces generation and debugging costs, with execution risks that can compound as more mechanisms are generated. To address those challenges, we introduce Harness Primitives, reusable harness mechanisms with clear application scope and composition contract mined from failed task trajectories. Based on Harness Primitives, we propose STITCH, a framework that Selects suitable primitives given Task Information and compiles them into Task-speCific Harnesses at test time. This separation enables task-specific harnesses without generating or repairing mechanism code at test time. Extensive experiments demonstrate that STITCH not only improves harness adaptability and robustness, but also scales with the primitive library size, boosting task success rates by up to 12 points over fixed harness baselines, surpassing human-designed harnesses like Codex CLI while maintaining a minimal test-time harness composition overhead of only 2.7%, 638 times more efficient than generating task-specific harnesses from scratch. Ultimately, our work demonstrates that building task-adaptive harnesses can be beneficial for completing diverse tasks and that building reusable primitives can be a promising path towards this goal.

cs.AI↗