arXiv subjects

Haoran Yu

Publications and source records attributed to Haoran Yu.

At least 37 records · Page 2Linked to original sources

arXiv2026

Phantom Guardrails: When Self-Improving Agent Harnesses Fix Failures That Never Happened

Self-improving AI agents are designed to learn from their mistakes. We show they can also hallucinate mistakes that never happened. We study this failure mode in automated harness optimization, where an LLM-based proposer edits an agent's scaffold, including prompts, parsers, filters, validators and guardrails, to eliminate observed failures. But this process rarely asks first: was there a real failure to fix? We introduce the Counterfactual Fabrication Lab, a deterministic micro-lab where the correct action is known: do nothing. The lab plants a candidate guardrail for a failure class that provably never occurs, presents only legal episodes, and uses a byte-exact oracle to check every cited violation. The proposer behaves as expected on real violations and abstains on featureless legal input. Yet when the legal input contains a harmless pattern resembling a familiar game rule, it invents a failure: in 15/60 runs, versus 0/60 on featureless input, it enables the nonexistent-rule guardrail and cites a violation the oracle refutes. The effect is structured, not indiscriminate. In single-shot proposals it appears only when three conditions coincide: a rule-shaped pattern, an open-ended rule set and an instruction that presupposes failures. Removing any of these conditions eliminates the fabrication. Because the invented guardrail changes no true outcome and cannot improve an already-perfect suppression score, the phenomenon is neither reward hacking nor over-refusal. It is a phantom guardrail: a fix for a failure that never happened, invisible to suppression-only acceptance. Inside an add-only accept loop it re-enters even without the failure-presupposing instruction, the loop's keep-adding role supplying the demand the instruction supplied in single shot, and once in it stays. We present the Counterfactual Fabrication Lab for measuring fabricated failures in self-improving agent harnesses.

cs.CR↗

arXiv2026

A GPU-Accelerated Framework for Multi-Attribute Range Filtered Approximate Nearest Neighbor Search

Range-filtered approximate nearest neighbor search (RFANNS) is increasingly critical for modern vector databases. However, existing solutions suffer from severe index inflation and construction overhead. Furthermore, they rely exclusively on CPUs for the heavy indexing and query processing, significantly restricting the throughput due to the limited memory bandwidth and parallelism. In this paper, we present Garfield, a GPU-accelerated framework for multi-attribute range filtered ANNS that overcomes these bottlenecks through designing a lightweight index structure and hardware-aware execution pipeline. Garfield introduces the GMG index, which partitions data into cells and builds local graph indexes. It guarantees linear storage and indexing overhead by adding a constant number of cross-cell edges. For queries, Garfield utilizes a cluster-guided ordering strategy that reorders query-relevant cells, enabling a highly efficient cell-by-cell traversal on the GPU that aggressively reuses candidates as entry points across cells. To handle datasets exceeding GPU memory, Garfield features a cell-oriented out-of-core pipeline. It dynamically schedules cells to minimize the number of active queries per batch and overlaps GPU computation with CPU-to-GPU index streaming. Extensive evaluations demonstrate that Garfield reduces index size by 4.4x, while delivering 119.8x higher throughput than state-of-the-art RFANNS methods.

cs.DB↗

arXiv2026

Operational Reframing and Approval-Framed Delegation in Multi-Agent LLM Safety

Safety evaluations of multi-agent LLM systems often compare a direct prompt with a planner-executor pipeline and report the difference as a single "pipeline effect." We argue that this aggregate is difficult to interpret because it conflates three mechanisms: harmful intent may be reframed as plausible operational work, the planner may refuse or transform the request, and the executor may act under delegation prompts implying prior approval. To separate these factors, we introduce a five-condition controlled contrast design, evaluated on 30 synthetic harmful scenarios and an exploratory external validation set from four agent-safety benchmarks using LLM-judged compliance. Our results show that aggregate pipeline safety is not a stable architectural property. Operational reframing is the most portable risk signal, increasing compliance for GPT, Gemini, and DeepSeek across both scenario sets, while Claude is comparatively resistant. Planner behavior can offset this risk mainly through refusal; however, when the planner produces executable steps, the executor may become more compliant than under the direct operational baseline. Approval-framed delegation is sensitive to prompt design, model pairing, and scenario source, and a skeptical executor prompt sharply reduces compliance. Raw-direct model rankings can also mispredict deployed planner-executor behavior. Gemini is safest under raw direct prompts in the primary set yet shows the largest amplification with a Claude planner, rising from 8.9 percent to 38.9 percent compliance. GPTs near-zero aggregate pipeline effect instead hides a reframing increase canceled by planner refusal. These findings suggest that multi-agent safety evaluations should report reframing, planner behavior, delegation framing, and model pairing separately before attributing failures to architecture itself.

cs.AI↗

arXiv2026

Stellar feedback drives the baryon deficiency in low-mass galaxies

Stellar feedback, as a key process regulating the baryon cycle, is thought to greatly redistribute baryonic material inside and outside the dark matter halos (DMHs), however the observational evidences are lacking. Through stacking analyses of ~400,000 galaxy spectra from Dark Energy Spectroscopic Instrument (DESI), we find star formation driven cool outflows in Mg II absorption line. Assuming only gravity acts on the launched gas, our calculations reveal that outflows from low mass galaxies ($M_*<10^{10}\,\rm M_\odot$) are capable of escaping beyond the DMHs, which aligns well with our finding in the circumgalactic medium (CGM) absorption along the minor-axes of galaxies using background quasars. This research offers indirect evidence that stellar feedback drives the low baryon retention rate in low-mass haloes, implicating that baryonic processes within galaxies are connected with the diffuse matter beyond the DMHs.

astro-ph.GA↗

arXiv2026

Cooler Phases of the Circumgalactic Medium Are More Centrally Concentrated: Constraints from Multiphase Absorption Lines

We present a systematic study of the multiphase circumgalactic medium (CGM) around galaxies and quasars, traced by Ca II $λ\lambda3934,3969$, Mg II $λ\lambda2796,2803$, and C IV $λ\lambda1548,1550$, using the Year 1 dataset from the Dark Energy Spectroscopic Instrument. These three doublets trace CGM gas across a range of temperatures, from cold to warm phases, and we employ a stacking technique to measure the corresponding absorption signals using background sources. We show that CGM structure is strongly phase-dependent: ions tracing progressively cooler gas exhibit increasingly steep radial profiles in equivalent width ($W_i$). These trends are broadly consistent with predictions from cosmological simulations, supporting a phase-stratified CGM in which cooler gas is more centrally concentrated. Specifically, halos of emission-line galaxies exhibit a strong radial transition from cool to warm gas, whereas halos of quasars show a more uniform distribution, likely regulated by active galactic nuclei feedback; in contrast, the cold gas traced by Ca II in low-redshift galaxies is tightly confined to inner regions. We further demonstrate that the radial scaling $W_i \propto D^α$ is primarily set by host stellar mass, particularly for the cool-phase medium, suggesting efficient heating processes in massive halos. By jointly leveraging multiple absorption tracers from observations and simulations, we map the CGM from cold to warm phases and place new constraints on the baryon cycle governing galaxy evolution.

astro-ph.GA↗

arXiv2026

ChainCaps: Composition-Safe Tool-Using Agents via Monotonic Capability Attenuation

Tool-using agents increasingly operate in open-ended deployment environments, where they compose file systems, web APIs, code interpreters, and enterprise services at runtime. This creates a safety gap in tool composition: an agent can satisfy every per-tool permission check and still produce an unsafe end-to-end effect, such as reading a confidential document, summarizing it, and sending the summary to an external endpoint. We call this failure mode permission laundering. ChainCaps addresses it with a runtime rule: every value carries a sink-specific capability budget, and tool composition propagates budgets by intersection. A value can preserve or lose authority as it moves through a tool chain, but it cannot gain new authority through composition. We implement ChainCaps as a transparent MCP proxy that requires no changes to the agent or tool servers. On 82 tasks across five frontier models from three providers, ChainCaps reduces attack success rate from 25-68% to 0-4.8% while preserving 96-100% benign completion. In replay experiments, it also outperforms scalar-IFC and per-function-isolation baselines. Manifest quality is the dominant deployment bottleneck: expert manifests reach 100% attack blocking, while naive manifests fall to 27.3%. Our claims are limited to explicit-flow composition safety under trusted manifests and proxy-visible data movement, a practical gap in deployed tool-using agents today.

cs.CR↗

arXiv2026

Beyond Simpson's Paradox: A Cascade of Confounders in AI Agent Pull-Request Co-Authorship

Pooled across five AI coding agents, pull requests (PRs) with a human Co-Authored-By trailer merge less often than purely-autonomous ones (53.8% vs. 79.8%) -- yet this aggregate finding is a textbook Simpson's Paradox. Stratifying 33,596 PRs from the AIDev dataset by agent identity reverses the conclusion: Copilot and Devin show large positive within-agent gaps (+41.2 and +33.5 pp, both p<0.001), while Cursor, Claude Code, and Codex show small effects whose cross-sectional 95% CIs span zero. The paradox is driven entirely by agent composition: Codex, which dominates 64.9% of the dataset, achieves high merge rates while rarely using co-authorship. But Simpson's Paradox is only the first layer of a cascade of confounders: within-repo controls eliminate Devin's gap (+33.5 to +1.6 pp, p=0.73); a commit-count control further halves Copilot's within-repo gap (+36.2 to +24.4 pp); restricted to multi-commit PRs, the Copilot within-repo effect dissolves to +4.8 pp (p=0.59). No agent retains a clear co-authorship effect once both repository selection and PR structure are controlled. Our findings caution against reporting agent-pooled statistics without stratification and demonstrate that cross-sectional co-authorship associations are largely selection and PR-structure artefacts rather than evidence of a causal benefit.

cs.SE↗

arXiv2026

Habituation at the Gate: Rising Approval and Declining Scrutiny in Human Review of AI Agent Code

As AI coding agents (e.g., GitHub Copilot, Devin, OpenAI Codex, Cursor) submit pull requests to open-source repositories at scale, a key question arises: do human reviewers gradually lower their scrutiny for AI-generated code over time? We conduct a longitudinal within-reviewer analysis using the AIDev dataset, studying 400 repeat reviewers who collectively submitted 11,429 reviews over a seven-month observation period. Comparing each reviewer's early and late review episodes, we observe a population-level shift in approval rate from 30.1% to 36.8% (Wilcoxon signed-rank p < 10^{-6} on paired shifts). Pooled by within-reviewer experience decile, the cumulative gap reaches +14.5 pp from first to tenth decile. This shift is experience-driven (persists after controlling for calendar time), agent-specific (human PR approval rates decline over the same period), and not explained by PR difficulty (median PR size is flat). However, review latency increases rather than decreases (+3.5x), while inline comment volume decreases (-22%, p=0.0014), suggesting reviewers spend more time in queue but less time actively inspecting code. The combination of rising approval, declining comment effort, and increasing queue time is most consistent with reflexive habituation under growing workload rather than rational trust calibration alone.

cs.SE↗

arXiv2026

When Safe Skills Collide: Measuring Compositional Risk in Agent Skill Ecosystems

LLM agents increasingly rely on community-contributed skills that expand an agent's operational capability set. We study a core safety problem in agentic AI systems: whether individually safe skills can compose into unsafe installed skill sets. We present SkillReact, a compositional security measurement framework with three components: a deterministic static-composition benchmark, a two-rater LLM-assisted human-adjudication pipeline, and an action-based exploitability harness. On 1,520 ClawHub skills, 651 pass individual inspection and form 211,575 pairs; the benchmark flags 22.25% of these as structural candidates. We treat this raw rate as a recall-oriented scanner ceiling and calibrate it against human judgment: in a pattern-stratified audit, roughly one in five flagged pair-pattern hits survives as a real compositional risk (population-weighted validity 18.2%, our headline result), implying about 14K genuine risk memberships in a single registry that per-skill scanning misses by construction, since every pair is individually safe. An action-based harness then probes when these candidates become model-issued tool calls, and finds realization gated by host-model disposition: on an anchor-conditioned dropper subset, Haiku-4-5 issues the dropper-stage tool call on all 39 direct-prompt trials (36 of them the full download-then-execute chain, 3 download-only), Opus-4-7 stops at the download, and Sonnet-4-6 refuses outright. A control that holds the request fixed and varies only the installed skills finds compliance highest with no skills installed: a composition fixes which capabilities are reachable, while the host model decides whether to use them. Together these motivate install-time compositional checks and capability isolation as complements to per-skill scanning.

cs.SE↗

arXiv2026

First Statistical Detection of MgII-traced Cool Gas Outflows with JWST Towards Cosmic Dawn

Galactic-scale outflows are a crucial component of galaxy evolution, yet their properties in the early universe remain poorly constrained. We present the first statistical investigation of \ion{Mg}{2}-traced cool gas outflows in galaxies spanning a wide cosmic timeline from $z \approx 1$ to $z > 6$ (with sample coverage extending to $z \sim 10$). Using thousands of public JWST/NIRSpec spectra, we employ a signal-to-noise weighted spectral stacking technique on the \ion{Mg}{2} $λ\lambda2796, 2803$ absorption doublet. We robustly detect blueshifted \ion{Mg}{2} absorption in nearly all stellar mass and redshift bins, with the exception of the lowest-mass systems at $z \sim 1-2$. The outflow equivalent width exhibits a positive correlation with stellar mass ($M_*$) at all epochs, with the fitted slope of $1.21 \pm 0.35$. Our work provides the first statistical constraints on \ion{Mg}{2}-traced cool outflows in the low-mass ($M_* \lesssim 10^{9.5} \, \mathrm{M}_\odot$), high-redshift ($z > 3$) regime. We also find that the outflow velocities generally remain below the host halo escape velocities, consistent with a galactic fountain scenario. The consistency of the stellar mass-outflow equivalent width relation across $z \sim 2-6$ suggests a persistent, unevolving feedback mechanism governing the baryon cycle towards cosmic dawn.

astro-ph.GA↗

arXiv2025

InstructUDrag: Joint Text Instructions and Object Dragging for Interactive Image Editing

Text-to-image diffusion models have shown great potential for image editing, with techniques such as text-based and object-dragging methods emerging as key approaches. However, each of these methods has inherent limitations: text-based methods struggle with precise object positioning, while object dragging methods are confined to static relocation. To address these issues, we propose InstructUDrag, a diffusion-based framework that combines text instructions with object dragging, enabling simultaneous object dragging and text-based image editing. Our framework treats object dragging as an image reconstruction process, divided into two synergistic branches. The moving-reconstruction branch utilizes energy-based gradient guidance to move objects accurately, refining cross-attention maps to enhance relocation precision. The text-driven editing branch shares gradient signals with the reconstruction branch, ensuring consistent transformations and allowing fine-grained control over object attributes. We also employ DDPM inversion and inject prior information into noise maps to preserve the structure of moved objects. Extensive experiments demonstrate that InstructUDrag facilitates flexible, high-fidelity image editing, offering both precision in object relocation and semantic control over image content.

cs.CV↗

arXiv2025

Central Concentration and Escape of Ionizing Photons in Galaxies at the Epoch of Reionization

Compact, low-mass galaxies with strong nebular emission are considered promising candidates for efficient ionizing photon production and escape. We present a spatially resolved analysis of 189 galaxies at redshifts $z \sim 6.7-7.6$ in JADES GOODS-N and GOODS-S fields and selected via JWST/NIRCam F410M filter. By employing annular photometry and spectral energy distribution fitting across rest-frame UV to optical wavelengths, we investigate the internal structure of star formation, ionizing photon production and escape, as well as the resolved star formation histories within these galaxies. We find that these galaxies exhibit compact, centrally concentrated, and bursty star formation, especially in lower-mass systems ($\log(M_*/{\rm M_{\odot}}) <9.0$). The central regions of them display extreme [OIII]+H$β$ equivalent widths ($>$1000 Å), high ionizing photon production efficiencies ($ξ_{\text{ion}} \sim 10^{25.6}$ Hz erg$^{-1}$), steep UV slopes ($\sim -2.3$), and elevated escape fractions ($f_{\text{esc}} > 0.08$), with all these properties peaking in the inner regions. These findings reveal outside-in growth and rising star formation histories at $z\sim 7$, with the central regions of them playing a pivotal role in driving cosmic reionization.

astro-ph.GA↗

arXiv2025

The Cosmic Evolution and Spatial Distribution of Multiphase Gas associated with QSOs

We investigate the multi-phase gas surrounding QSOs traced by 33 absorption lines (e.g., Ly$α$, C\,\textsc{iv}, Fe\,\textsc{ii}, Mg\,\textsc{ii}, etc.) in the stacked spectra of background sources, using the early data release from the Dark Energy Spectroscopic Instrument. Our analysis reveals that the equivalent width ($ W $) of metal absorption lines decreases with increasing redshift, following an overall trend described by $W \propto (1+z)^{-4.0\pm 2.7}$. Different species that trace multi-phases of QSO-associated gas exhibit distinct evolutionary patterns. Additionally, the $ W $ of these absorption lines decreases with distance ($D$) from QSOs, which can be effectively characterized by a two-halo model. Compared to the projected two point correlation function of galaxies at similar redshifts, low-ionization ions exhibit similar clustering scales, while high-ionization ions show a significantly more extended spatial distribution. We also find that $W_{\text{FeII}}/W_{\text{MgII}}$ increases towards lower redshifts, which can be attributed to evolving star formation histories and/or changes in initial mass function for galaxies. By leveraging multiple absorption tracers, we conduct the first comprehensive investigation of diffuse, multiphase gas from the circumgalactic medium to cosmological scales, offering new insights into baryon cycles and the transport of metals throughout cosmic time.

astro-ph.GA↗

arXiv2025

Potential-Driven Metal Cycling: JADES Census of Gas-Phase Metallicity for galaxies at 1 < z < 7

The gravitational potential is established as a critical determinant of gas-phase metallicity (12+log(O/H)) in low-redshift galaxies, whereas its influence remains unconfirmed at high redshifts. We investigate the correlation between gas-phase metallicity and effective radius ($R_{\rm e}$) for a sample of galaxies with redshifts ranging from 1 to 7, drawn from JADES (JWST Advanced Deep Extragalactic Survey) Data Release 3. We calculate the metallicities using four strong-line methods: ${\rm N2S2Hα}$, ${\rm R23}$, ${\rm N2}$, and ${\rm O3N2}$, respectively. After taking out the evolution of size, we find that the offsets of mass-size relation ($Δ\log R_{\rm e}$) are significantly negatively correlated with the offset of mass-metallicity relation ($Δ\log({\rm O/H})$) for the four metallicity tracers. Regardless of the metallicity tracer used, we obtain Spearman rank $p-$values much less than 0.01, rejecting the null hypothesis that the observed correlation is statistically nonsignificant and attributable to random chance. This is also true for galaxies with $z>3$, with $p-$values less than 0.05 for the four metallicity tracers. We for the first time find evidence of size playing a key role in determining gas-phase metallicity towards cosmic dawn, suggesting that the gravitational potential influences their material-exchange processes with the surrounding environment at very early universe.

astro-ph.GA↗

arXiv2025

The $M_{\rm BH}-M_\star$ Relation of the hyperluminous Dust-obscured Quasars up to $z \sim 4$

Hot dust-obscured galaxies (Hot DOGs) are a rare population of hyperluminous dust-obscured quasars discovered by the Wide-field Infrared Survey Explorer (WISE) all-sky survey. The heavy circumnuclear dust obscuration allows only a small amount of scattered light from the obscured quasar to escape, enabling the decomposition of the stellar component from the total flux. The presence of scattered light enables the redshift of the source and the properties of the black hole to be obtained from SDSS and SDSS-related literature. From WISE and SDSS data, we select 11 hyperluminous Hot DOGs at $z=1.5-3.7$ with bolometric luminosities $L_{\rm bol} \gtrsim 10^{47}\,\mathrm{erg \ s^{-1}}$. We investigate the $M_{\rm BH}-M_\star$ relation in these sources using Bayesian spectral energy distribution (SED) fitting or with extra constraints from \textit{Hubble Space Telescope} (HST) image decomposition. Stellar masses are successfully derived for eight Hot DOGs. We find high Eddington ratios $λ_{\rm Edd}$ in these Hot DOGs, with the median value of 1.05 and the maximum value close to 3. The super-Eddington accretion may be associated with the overdense environments of Hot DOGs. We find no significant differences in the $M_{\rm BH}/M_\star$ of these Hot DOGs compared to the local relation, suggesting that these dust-obscured quasars are the progenitors of massive early-type galaxies. We speculate that the subsequent evolution of Hot DOGs may be significantly influenced by AGN feedback and remain on the local relation.

astro-ph.GA↗

arXiv2025

Central Velocity Dispersion being the Primary Driver of Abundance Patterns in Quenched Galaxies

The element abundances of galaxies provide crucial insights into their formation and evolution. Using high-resolution IFU data from the MaNGA survey, we analyze the central spectra (0-0.5 $R_{\rm e}$) of 1,185 quenched galaxies ($z = 0.012-0.15$) to study their element abundances and stellar populations. We employ the full-spectrum fitting code {\tt alf} to derive stellar ages and element abundances from synthetic spectra and empirical libraries. Our key findings are: (1) Central velocity dispersion ($σ_*$) is the most effective parameter correlating with (relative) element abundances, especially [Na/Fe], [Mg/Fe], [C/Fe], and [N/Fe], outperforming $M_\ast$ and $M_\ast/R_{\rm e}$. (2) When binned by $σ_*$, the relative abundances of Na, Mg, C, and N remain stable across different formation times ($T_{\rm form}$), suggesting these elements are primarily influenced by the burstiness of star formation (traced by $σ_*$) rather than prolonged evolutionary processes. (3) Fe and Ca show little variation with $σ_*$, indicating weaker sensitivity to $σ_*$-driven processes. However, $T_{\rm form}$ has a global influence on all elements, contributing to their overall chemical evolution, albeit secondary to $σ_*$ for most elements. These results support the primary role of $σ_*$ in shaping the abundance patterns, likely stemming from the connection between central massive black holes and possibly dark matter halos, which influences the burstiness of star formation histories.

astro-ph.GA↗

arXiv2025

Dominant Role of Coplanar Inflows in Driving Disk Evolution Revealed by Gas-Phase Metallicity Gradients

Using spatially resolved spectroscopic data from the MaNGA sample, we investigate the parameters influencing the radial gradients of gas-phase metallicity ($\nabla\log(\mathrm{O/H})$), to determine whether disk formation is primarily driven by coplanar gas inflow or by the independent evolution of distinct regions within the disk. Our results show that $\nabla \log(\mathrm{O/H})$ strongly correlates with local gas-phase metallicity at a given stellar mass, with steeper gradients observed in metal-poorer disks. This trend supports the coplanar gas inflow scenario, wherein the gas is progressively enriched by in situ star formation as it flows inward. In contrast, the radial gradient of stellar mass surface density shows very weak correlations with $\nabla \log(\mathrm{O/H})$, which is inconsistent with the independent evolution mode, where gas inflow, star formation, and metal enrichment occur independently within each annulus of the disk. Furthermore, we find that $\nabla \log(\mathrm{O/H})$ is also closely correlated with an indicator of local gas turbulence $σ_{\mathrm{gas}}/R_{\mathrm{e}}$, highlighting the competing roles of turbulence and coplanar inflow in shaping metallicity gradients. Our results provide indirect observational evidence supporting coplanar gas inflow as the driving mechanism for disk evolution.

astro-ph.GA↗

arXiv2025

The circumgalactic medium traced by Mg II absorption with DESI: dependence on galaxy stellar mass, star formation rate and azimuthal angle

Understanding the circumgalactic medium (CGM) distribution of galaxies is the key to revealing the dynamical exchange of materials between galaxies and their surroundings. In this work, we use DESI EDR dataset to investigate the cool CGM of galaxies ($0.3 10.0$, the EW(Mg II) at the minor axis is largely suppressed with respect to low mass galaxies. This suggests that the competing processes, such as stellar feedback and gravity, play a key role in shaping the distribution of outflowing gas.

astro-ph.GA↗