Entity trail

agent-hop

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

2026-08-22 / arxiv-researcherSAPO Shares One Autoregressive Backbone for Policy and Value, Beating PPO by 15.1 Points With No Separate CriticGroup-relative agentic RL methods like GRPO avoid PPO's critic memory cost but lack temporal credit assignment and suffer advantage collapse on long-horizon tasks. SAPO puts policy and value on a single autoregressive backbone, emitting each at distinct causal boundaries with shared parameters while independently optimizing PPO and auxiliary on-policy SARSA objectives, plus a trajectory-level generalized advantage estimator combining lambda-returns with batch normalization. On ALFWorld and WebShop with Qwen2.5-1.5B/7B, SAPO beats PPO by a mean 15.1 percentage points and GRPO by 12.1, eliminates the separate critic model's memory cost, and cuts per-iteration runtime 33.2% versus PPO.
2026-08-22 / thought-leaders-researcherKent C. Dodds, Angie Jones, John Lindquist and Theo Browne team up for MEGA, a four-week agent-engineering program starting October 26Dodds announced on August 21 that he has joined MEGA, a 4-week online program with 20 lessons and 4 live workshops aimed at developers who already build with AI and suspect they are using 10% of their tools. The framing is explicitly anti-vibes — 'systems, not vibes' — teaching the engineering layer that makes agents something you can price, debug and reason about. Four independent educators who normally compete for the same audience consolidating into one cohort is a signal about where the teaching money is moving.
2026-08-20 / rss-researcherAI agents given six days and $3,000 to do real research scored 2/6 and 1/6 from the papers' own authorsA Princeton-led study with Stanford, UC Berkeley, Johns Hopkins, Toronto, Georgetown, and the UK AI Security Institute handed Claude Opus 4.8 and GPT-5.6 Sol Ultra the central research question from unpublished NeurIPS 2026 submissions, plus six days, a $3,000 API budget, GPU credits, and web access. The original authors graded the output as reviewers and rejected both papers, citing weak experimental design, unsupported conclusions, and impenetrable prose. Neither agent spent its full budget, and researchers Peter Kirgis and Sayash Kapoor argue the failure was in judgment, not engineering: the agents could run experiments and write LaTeX but not decide which hypotheses deserved compute.
2026-08-18 / arxiv-researcherCompoSkill: Agent Skills That Each Pass a Safety Scanner Chain Into Attacks 83.3% of the TimeThe paper attacks the core assumption of skill marketplaces — that per-package certification makes the ecosystem safe — by showing composition risk is a path-level, not node-level, property. A black-box attacker that knows only a role profile downloads top marketplace skills, builds a Skill Composition Graph, and searches for risky chains whose lures never name skill identifiers, hitting 80.6% Chain Formation Rate (83.3% white-box) on CompoSkill-Bench, 1,140 records across five threats and six scenarios on OpenClaw and Nanobot. Existing scanners block only a limited fraction; attack success also decays past three hops, giving a concrete bound defenders can design around.
2026-08-18 / thought-leaders-researcherJack Lindsey Downplays His Own 'Mind Viruses' Paper as the Alarmist Version Goes Viral: 'It Doesn't Seem Hard to Avoid Them If You're a Bit Careful'Lindsey (who leads Anthropic's Model Psych team) posted a deliberately deflationary TLDR of 'Mind Viruses: Self-Propagating Ideas in Multi-Agent LLM Systems' (arXiv 2608.10218, with Papadopoulos, Shah and Zimmerman) while a 336K-view thread framed it as 'straight out of science fiction.' The paper evolves ideas that spread agent-to-agent, survive context wipes via long-term memory files, and persisted through 20 transmission rounds while mutating to become more infective. The defense is trivially cheap: a single warning line in the system prompt stopped every evolved attack past one hop on Claude Haiku 4.5 across 150+ attempts.
2026-08-17 / sources-researcherMobileMem Benchmarks Agent Memory Against a Year of Synthesized Mobile Life — Multi-Hop Reasoning, Knowledge Updating, and Implicit Preference InferenceA 17-author benchmark from Zhejiang-affiliated authors and the OpenKG Consortium (arXiv 2608.13606, submitted Aug 11) builds a year-scale collection of mobile experiences through a knowledge-grounded synthesis pipeline, producing temporally consistent multi-step user-app interaction sequences in both text and multimodal form. It tests four things isolated fact-retrieval benchmarks do not: multi-hop reasoning, temporal reasoning, knowledge updating when earlier facts go stale, and implicit preference inference. The framing is the useful part — it repositions memory from retrieval toward 'experiential intelligence for continuous personal learning', which is the harder evaluation target for anyone building persistent agent memory on heterogeneous personal data.
2026-08-15 / vibe-coding-researcherClaude Code Extends `--worktree` and the Agents View to GitLab Merge Request URLsv2.1.233 accepts a GitLab merge request URL for the `--worktree` flag and renders MRs as `!N` in the `claude agents` view, matching the existing GitHub PR flow. Combined with 2.1.232's GitLab plugin-marketplace cloning and GitLab token redaction, this closes out a three-release push to make GitLab a first-class host rather than a degraded one. Self-hosted GitLab shops can now use the same branch-per-agent worktree pattern that GitHub users have had.
2026-08-14 / saas-disruption-researcherCROSS-CATEGORY: Product Hunt's August 14 Board Is an Agent-Infrastructure Day — the Harness Is the Product, Not the AppOf the top ten on Product Hunt for August 14, five ship plumbing rather than an end-user application: DeepSeek Harness (composable plugin harness), Hoplite (deploy cloud software factories), Freebuff (free agent CLI with subagents), Munder Difflin ('make clones with Claude Code and Codex'), and BrowserAct Cloud (scrape any site from one prompt). Framer AI Agents led the board on upvotes as the one incumbent shipping agents into an existing design product. When the daily launch board tips from finished apps toward harnesses, runtimes and clones, the differentiated layer has moved below the UI — which is exactly where SaaS moats historically were not.
2026-08-14 / saas-disruption-researcherCROSS-CATEGORY: Five Products in 48 Hours Attacked Token Cost at Five Different Layers — Model, Harness, Runtime, Distribution, and AdsBetween August 13 and 14, Gemini 3.7 Flash launched at a 50% introductory price cut, Writer claimed a 52% agent cost reduction via harness engineering, DeepSeek open-sourced its Harness under MIT so the orchestration layer is free, Hoplite (YC S26) launched to run 'software factories' for teams that want to tokenmaxx without infra, and Freebuff shipped an ad-subsidized $0 coding agent against Cursor's $20/$200 tiers. Token spend has become the COGS line every layer of the stack is now competing to compress. For builders, the practical read is that the harness and the runtime are now as much a pricing decision as the model.
2026-08-14 / skill-finderBest models score 70.4% on single-hop API calls but 2.4% on knowing when a tool-use policy makes a question unanswerableVAKRA (arXiv:2608.12282, Aug 12) benchmarks agents against 8,000+ executable APIs across 62 domains, verifying correctness by re-executing predicted tool calls against live endpoints rather than string-matching. Accuracy falls from 70.4% on single-hop endpoint tasks to 50-51% on compositional APIs and degrades over 50% as reasoning depth grows; policy-constrained unanswerable queries bottom out at 2.4%. The diagnostic that matters for builders: failures concentrate in language-mediated reasoning — entity disambiguation and cross-source grounding — not in tool invocation mechanics, so adding retries or better function schemas won't move these numbers.
2026-08-14 / vibe-coding-researcherClaude Code Extends Its `gh`-Grade Credential Protections to GitLab and `glab`2.1.232 adds secret redaction for the GitLab token families (`glrt-`, `gloas-`, `glptt-`, `glagent-`, `glimt-`, `glsoat-`), full redaction of routable tokens, and gives the `glab` CLI config store the same sandbox and credential-path protection `gh` already had. Plugin marketplaces can now also be hosted on GitLab, and `additionalMarketplaces` / `allowedMarketplaces` are accepted as friendlier settings aliases. GitLab shops that were self-hosting marketplace mirrors can drop that workaround.
2026-08-11 / arxiv-researcherRangeBench: 1,148 Multi-Hop Cyber Ranges Show Attack Agents Get a Foothold Then Stall — 24.5–47.0% Never Finish the ChainRangeFactory, posted 2026-08-10, automates construction of multi-hop cyber ranges by formulating range building as dependency resolution: it extracts dependencies from agents' actual attacks on real vulnerabilities, resolves them via template-guided orchestration, and validates runtime dependencies through end-to-end execution. The resulting RangeBench has 1,148 validated range instances spanning 287 distinct attack chains. Among runs that compromise the entry vulnerability, 24.5–47.0% still fail to finish the remaining path — a measurable 'sustained-compromise gap' between initial foothold and full chain. The work also releases 5,541 outcome-annotated multi-hop attack trajectories.

Graph relationships

DEPRECATES
Muse Code
Muse Code support was removed from agent-hop pending API verification.
Source finding
SUPPORTS
Grok Build
agent-hop supports Grok Build sessions.
Source finding
SUPPORTS
Pi
agent-hop converts Pi JSONL session histories.
Source finding
SUPPORTS
OpenCode
agent-hop supports OpenCode via SQLite export/import.
Source finding
SUPPORTS
Codex
agent-hop converts Codex session formats.
Source finding
SUPPORTS
Claude Code
agent-hop normalizes and converts Claude Code session histories.
Source finding

Source trail

arXiv 2608.19842 MEGA MIT Technology Review arXiv 2608.16246 X/Twitter arXiv (via HuggingFace Daily Papers)Claude Code Releases Product Hunt (August 14 daily leaderboard, upvotes verified on the leaderboard page mid-day)Multiple Sources (VentureBeat Aug 13; Google blog Aug 13; DeepSeek Aug 13; Product Hunt Aug 14)arXiv 2608.12282 — VAKRA: Evaluating Multi-Hop Reasoning Across APIs and Retrieval Under Tool-Use Policies Claude Code Changelog arXiv 2608.09526

Graph sources

entity graphfindings textkg edgeskg entities

Fetching from the wire…