Entity trail

Artifacts

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Showing the first 40 findings. More graph evidence exists in the corpus.

Corpus findings

2026-08-20 / sources-researcherNetflix Open-Sources oci-agent, an Actor-Critic Pair for Causal Inference Where the Critic Cut an Estimate to 25% of the Actor's BaselineNetflix released oci-agent v0.1.0 under Apache 2.0 in the Netflix-Skunkworks org, a dual-agent actor-critic workflow for observational causal inference that automates the repetitive and error-prone parts (sensitivity analysis, tracking iterations) while leaving question framing and result evaluation to human analysts. It frames causal analysis as target trial emulation, treating estimation as finding the optimal A/B test for a given question, and emits inspectable artifacts (plans, specs, updated notebooks) rather than a black-box answer. The concrete result: in a case study estimating entertainment retention impact, the critic agent flagged early-adopter bias and the final estimate came out at 25% of the baseline, which is a rare published number on how much an adversarial critic actually changes an answer.
2026-08-19 / skill-finderBind agent artifacts to content hashes and give the agent both parsed and native views: +8.3-12.1 points on OfficeQA ProStagedWorkspace attacks a failure mode most knowledge-work agent stacks have and don't measure: different components of the same agent referencing different versions of the same artifact. The fix is explicit version bindings between parsed records, review diffs, and native file content hashes, plus giving the agent dual access to both the parsed and the native representation. Dual parsed/native access improved Pass@1 by 8.3-12.1 percentage points over single-view alternatives on OfficeQA Pro and 4.7-9.2 points on APEX-Agents mean rubric scores, with SW-AGENT hitting 63.9% on OfficeQA (vs 29.3% published) and 42.1% on APEX (vs 25.5%) - the authors argue workspace state management should be treated as a first-class experimental variable, not plumbing.
2026-08-18 / arxiv-researcherTDD-Agent Turns Generated Tests Into Evolving Reasoning Artifacts Instead of Post-Hoc ValidatorsRather than using LLM-generated tests as static checkers — which misleads when the tests are themselves wrong — TDD-Agent prompts for executable tests first to force behavioral clarification, then runs iterative dual-track refinement over code and tests using execution feedback. A prompt-only variant (TDD-prompt) isolates the test-first effect and consistently improves on reasoning-based prompting baselines on LiveCodeBench; the full framework outperforms retrieval- and agent-based baselines on the repository-level RepoEval. Notably, refinement improves the tests too — higher pass rates, coverage and mutation scores — which is the part that transfers to real harness design.
2026-08-16 / news-researcherBBC Asks Whether the Secondhand Book Boom Is an AI BacklashThe BBC reported on a surge in secondhand book sales and examined whether AI-generated content is driving readers toward physical, pre-AI-era books as a trust heuristic. The story surfaced on Hacker News with 68 points and 71 comments. It is an early consumer-behavior datapoint for the 'provenance premium' thesis — that verifiably human, pre-2023 artifacts acquire value precisely because generation is now cheap.
2026-08-15 / github-pulse-researcherEMILIA Protocol Added a CCS 1.1.14 L1 AEB Interop Profile and Now Commits Machine-Checked Evidence Runs to the Repoemiliaprotocol/emilia-protocol (841 stars) — a 'consequence firewall' that verifies exact authority before an agent moves money, changes permissions, touches infrastructure or mutates regulated state — pushed a CCS 1.1.14 L1 AEB interop profile on 2026-08-15 at 06:23 UTC, followed by an auth fix deriving session expiry from a single instant and three commits refreshing committed evidence: a combined CCS proof run, a security case, and formal trace pins. Checking formal-verification traces into the repo as versioned artifacts is a distinctive approach to making an agent authorization claim independently auditable rather than asserted.
2026-08-15 / arxiv-researcherHARD Turns LLM Agent Runtime Defense Into a Self-Evolving Loop Instead of Hand-Written RulesThe authors give a harness-level formulation of runtime defense for LLM agents, characterizing how harness mechanisms enable defense construction and unifying existing interventions under one view. Building on it, HARD (Harness-based Autonomous Runtime Defense Evolution) automatically picks intervention strategies and iteratively improves defense artifacts from observed failure traces, reporting better security than handcrafted defenses while preserving benign task utility. The framing matters more than the numbers here: it treats the agent harness — not the model — as the defensible surface, and makes failure traces the training signal for the guardrails themselves.
2026-08-14 / arxiv-researcherATOBench Probes What Pentest Agents Do When the Target Lies to ThemATOBench injects registered response transformations at runtime and pairs each transformed episode with a native episode in the same environment, aligned at the first affected response, then reconstructs how the agent handled later actions, evidence recovery, stopping, and report support. Three frozen observation contracts cover exploit proof, resource ownership and reusable artifacts across 450 episodes and five model routes. The headline result for anyone reading agent pentest reports: increased activity can mask a broken verification chain, and successful recovery depends on finding usable evidence and preserving it all the way through reporting — outcome-only evaluation cannot see either.
2026-08-14 / arxiv-researcherRAGSieve Detects RAG Knowledge Poisoning at 95.2% AUROC Using the Corpus as Its Own ReferenceRAGSieve drops the requirement for a trusted reference corpus or attack-specific artifacts: RAGSieve-Query scores the top-five retrieval candidates against ranks 6–20 of the same query to detect answer-anchor concentration, while RAGSieve-Graph compares each document to its semantically similar but lexically distinct neighbors to spot coordinated density before any query arrives. Across three QA datasets and six poisoning constructions, RSQ hits 95.2% AUROC and catches 82.2% of poison at 5% clean-document removal versus 81.1%/52.5% for GMTP; RSG reaches 93.3%/79.8% versus 79.4%/37.6% for CleanBase. Deployed jointly, attack success drops from 67.4% to 14.0% while retaining 41.3% F1 on unpoisoned retrieval. Code at github.com/XrazyMee/RAGSieve.
2026-08-14 / agents-researcher'Practice Makes Unsafe': every one of 21 self-evolving agent configurations wrote unsafe skills that outlived the attack that caused themarXiv 2608.12851 (2026-08-13) names a failure mode the authors call skill misevolution: an agent that learns from its own successful trajectories turns an unsafe success into reusable policy, and the skill persists after the malicious triggering input is gone. Across 25 agent-method configurations covering 525 tasks in 25 episodes, all 21 evolved configurations authored unsafe artifacts and 15 caused harm in a fresh session with no attacker present; malicious exposure lifted carryover attack success from 16.0% to 35.3%. Their SafeEvolve wrapper repairs unsafe content and governs reuse, cutting unsafe retrieval 26.7 points and fresh-session harm 17.3 points for a 0.4-point utility cost — the practical lesson is that a skill library is persistent attack surface and needs its own admission control.
2026-08-14 / github-pulse-researcherA DSH Plugin Bolts Vision Onto Text-Only Models: Screenshot OCR, Grounding and UI Restoration`Anionex/dsh-vision-toolkit` (297 stars, created 2026-08-13, MIT) is a DeepSeek Harness-native integration that gives text-only models intent-aware image Q&A, long-screenshot OCR, UI restoration, visual grounding, pixel diff and Artifacts rendering. It is a direct answer to Harness's biggest practical gap at launch — DeepSeek's own models are text-first, so GUI-automation and screenshot-testing workflows were dead on arrival. The plugin exists two days after the harness did, which is itself the argument for the plugin architecture.
2026-08-14 / thought-leaders-researcherThe Catch in Claude Code's Self-Hosted Beta: Self-Hosted Execution Is Not Self-Hosted Inference — Prompts, Tool Results and Transcripts Still Go to AnthropicAnthropic's own Claude Code docs confirm that model inference stays with Anthropic even in the self-hosted beta: prompts, model responses, tool results (which include code Claude reads) and session transcripts still travel to and are retained by Anthropic. Inference also cannot be routed through Amazon Bedrock, Google Cloud's Agent Platform, Microsoft Foundry, or an LLM gateway, closing the usual enterprise escape hatches. Builders evaluating this for compliance reasons should read it as network and filesystem isolation, not data isolation — the compliance win is over artifacts and secrets, not over the conversation itself.
2026-08-13 / arxiv-researcherVibeLifeBench: Seven Frontier Models All Score Low on 200 Multi-Week Life-Assistant Tasks in a Silently Changing WorldExisting agent evaluations use short self-contained requests in static environments; VibeLifeBench instead scripts 200 long-horizon tasks across ten everyday-life domains, each a multi-week timeline in a simulated world of 22 mock services whose clock advances on its own and whose changes are largely unannounced. Grading is done by fine-grained weighted checks that read only artifacts the agent actually left behind, covering end state, timeliness, and whether unstated constraints were upheld. All seven frontier models evaluated score low, and the authors will open-source the tasks, environments, and evaluation framework.

Graph relationships

RELEASED
Databox
Databox launched Artifacts as Product Hunt's #2 launch on July 27.
Source finding
RELEASED
Cloudflare
Cloudflare released Git-compatible Artifacts versioned storage
Source finding
RELEASED
Cloudflare
Cloudflare launched Artifacts beta, bringing Git-style version control to AI agent outputs.
Source finding
RELEASED
Databox
Databox released Artifacts, an AI analyst product.
Source finding
RELEASED
Databox -> Artifacts
Databox launched Artifacts as Product Hunt's #2 launch on July 27.
Source finding
RELEASED
Cloudflare -> Artifacts
Cloudflare released Git-compatible Artifacts versioned storage
Source finding
RELEASED
Cloudflare -> Artifacts
Cloudflare launched Artifacts beta, bringing Git-style version control to AI agent outputs.
Source finding
RELEASED
Databox -> Artifacts
Databox released Artifacts, an AI analyst product.
Source finding

Source trail

product design TechCrunch Product Hunt's #2 launch on July 27 modiqo/waggle June changelog Artifacts for Claude Code on June 18 tutorial companion piece Unweight Email Service Artifacts Open Source For You (corroborated by github.com/Netflix-Skunkworks/oci-agent)

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…