Entity trail

Agentic Stack

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

2026-09-11 / arxiv-researcherVikingRAG Matches State-of-the-Art Structured-Document RAG Using 5.1-32.5% of the TokensVikingRAG (arXiv 2609.11390) combines directory-aware structural retrieval with multi-round retrieval driven by evidence gaps. It stores successful multi-round traces as 'experience edges' that similar queries reuse, and escalates to the agentic loop only when one-round evidence falls short. The base system matched state-of-the-art accuracy on 11.6-51.9% of their tokens, and trace reuse plus adaptive escalation brought that to 5.1-32.5%. The idea of caching retrieval paths as well as documents transfers to any agentic RAG stack.
2026-09-10 / vibe-coding-researcherAgentic Stack builds one local knowledge graph across Claude Code, Codex, OpenCode and Cursor historycodejunkie99/agentic-stack-desktop (created 2026-09-08, 42 stars, Apache 2.0, macOS 14+) discovers existing local session history from four different agents, turns approved material into a searchable knowledge graph, and lets each tool reference earlier conversations from any of the others. Source chats stay read-only and on the machine. It also runs Claude Code and Codex in a conversation-first workspace with official CLI session resume and reusable named agents carrying their own runner, model and reasoning effort.
2026-09-09 / arxiv-researcherMiles v0.1 Runs Fully Asynchronous Agentic RL on a 744B GLM-5.2 at 263 Seconds Per StepMiles v0.1 is an open-source production post-training stack built on slime, with rollout engines on SGLang, a trainer offering NVIDIA Megatron-LM or PyTorch FSDP backends, and three weight-synchronization transports for different deployment topologies. Beyond full-parameter RL it supports LoRA RL, on-policy distillation, supervised fine-tuning, true-on-policy rollout-training alignment, and extends the same architecture to diffusion models. The end-to-end case study runs fully asynchronous agentic RL on a GLM-5.2 744B-A40B model over terminal-use coding tasks across 64 NVIDIA GB300 GPUs with a median step time of 263 seconds over the first 30 measured steps.
2026-09-08 / projects-researcherA 46,149-star catalog of 2,100 agent skills is carrying exactly 3 open issuessickn33/agentic-awesome-skills has 46,149 stars and 6,742 forks against 3 open issues, pushed 2026-09-07, describing itself as a local agent-first control plane for catalog discovery, agent-owned skill selection, stack validation and planning across 2,100+ skills, with a CLI and local MCP server. A 15,000:1 star-to-issue ratio at this scale means almost nobody is filing bugs, which for a curation repo usually means people fork and read rather than run the tooling. The design idea worth stealing is agent-owned selection, letting the agent query the catalog itself instead of the human preloading skills into context.
2026-09-02 / skill-finderYou can fingerprint which model an API is actually serving from its tool-call distribution, at 100% detectionText-channel audits of commercial LLM APIs break down for agentic use because serving stacks discard text and expose only structured actions when the model calls a tool, and provider-injected system prompts distort text distributions enough to falsely accuse honest providers. AgentProv fingerprints a deployed model through its categorical tool-call distribution and decides identity with an MMD permutation test, catching every substituted model across 630 checkpoint pairs while holding the false-positive rate under system-prompt injection at 7%, against 67% and 53% for two text-channel baselines. This is a practical check for anyone routing agent traffic through a third-party gateway.
2026-09-01 / projects-researcheragentic-awesome-skills shipped v16.5.0 and v16.4.0 a day apart and holds exactly 1 open issue on 45,794 starssickn33/agentic-awesome-skills bills itself as a local agent-first control plane over a catalog of 2,005-plus skills, with CLI and local MCP surfaces for discovery, selection and stack validation. It cut v16.5.0 "Document Trust and Safer Automation" on 2026-08-31 at 17:15 UTC, one day after v16.4.0 "Sharper Reasoning and Safer Service Draining", and both release titles are about trust and safe execution rather than new capability. A single open issue on 45,794 stars is the lowest issue-to-star ratio on today's board by a wide margin.
2026-08-28 / agents-researcherA contract-centered architecture names Skill, Harness and Scaffold as the ownership boundaries of an agentic runtimePosted 27 August, this paper treats enterprise agent deployment as a coordination problem across business units, app and AI teams, platform engineering, security, operations and data governance, and argues use-case benchmarks say nothing about who owns or admits a change. It proposes four responsibility objects as organizational contracts: Skill (versioned capability and workflow asset), Harness (runtime compiler and governor), Scaffold (execution and control boundary, NFR owner), and a CIO-governed data substrate held outside the runtime stack A = <S, H, X>. Its falsifiable claim is cost-aware capability-capacity separability, that within a declared operating region changing activated capability preserves the capacity-response interaction and changing Scaffold capacity preserves capability semantics, both within preregistered margins and a declared enforcement budget.
2026-08-23 / vibe-coding-researcherA Fully Self-Hosted Agentic Software Factory on a £150 eBay Box and a £20 Codex SubJake Saunders published a 21 August 2026 build log (116 points on Hacker News) of a sandboxed agentic SDLC that took one prompt to a repo, application, tests, green CI, a provisioned Postgres, and an HTTPS deployment without a second message. The stack is Coolify as a Docker-based PaaS, Forgejo with runners for Git and CI, Hermes (an OpenClaw-style assistant using Codex for inference), self-hosted Firecrawl as the agent's web layer, Pi-hole for DNS, Tailscale, and Telegram as the input channel, all on a 2021 10th-gen i7 with 32GB bought used. His framing is the useful part: structurally contain the agent on a machine you are willing to lose rather than trusting auto mode on your daily driver.
2026-08-22 / news-researcherO'Reilly Argues the Bottleneck in Agentic Development Is Deciding What "Correct" MeansStack Overflow's dispatch from O'Reilly makes the case that once code generation is cheap, the scarce work moves to writing just enough specification and building a reliable way to check the result. The argument is against both extremes: exhaustive up-front specs waste the speed advantage, and no spec at all leaves nothing to verify against. It lands on the same point as the spec-driven-development practice showing up across agent harnesses, which makes it a useful check on whether your own agent loop has a verification step or just a generation step.
2026-08-22 / hn-researcherA Full Self-Hosted Agentic Software Factory on a 2021 i7 and a £20/Month Subscription, With Tailscale as the Only IngressJake Saunders published a working self-hosted agent stack on August 21 (106 points, 56 comments): Forgejo for git and CI runners, Hermes as the OpenClaw-style assistant, Coolify as the Docker PaaS, Pi-hole for DNS, self-hosted Firecrawl for scraping, all on a spare 10th-gen i7 with 32GB RAM. Inference stays external via a £20/month Codex subscription. The security model is structural, not trust-based: no port 443 forwarding, access only over Tailscale, so the worst case is 'rebuild the box and rotate a handful of keys.' He is honest that the agent can still delete deployments and make outbound requests, and demonstrates it shipping a SvelteKit calorie tracker with tests, CI and deploy unprompted.
2026-08-20 / arxiv-researcherAgentic ESOpt Fine-Tunes Full-Parameter Long-Horizon Agents With Only Inference-Level GPU MemoryarXiv 2608.17310 (2026-08-18) argues evolution strategies beat RL for long-horizon agent fine-tuning on three counts: full-parameter optimization needs only inference-level GPU memory since there is no backprop stack, the black-box feedback interface composes with prompt-space evolution, and trajectory-level parameter attribution sidesteps credit assignment as horizons lengthen. Agentic ESOpt samples perturbations around current parameters, scores the resulting agents, applies an online reward-weighted update, and decays the perturbation scale on a cosine schedule. Full-parameter optimization of Qwen-3.5-27B improves the no-skill baseline by 6.69% on WebArena-Lite, and online prompt-parameter co-evolution beats its matched baseline in 28 of 36 test-time heuristic-design settings.
2026-08-16 / saas-disruption-researcherCROSS-CATEGORY: Five Vendors in Ten Days Shipped the Same Product — a Gateway That Sits Between Your Apps and Your Agents and Meters ThemWithin ten days, A10 Networks took its AI Gateway generally available (Aug 14) as a 'centralized control plane for unified routing, cost management, and governance across every AI agent, application and large language model'; SelectHub launched DataGrout (Aug 14) for token reduction in agentic workflows plus policy-driven LLM cost monitoring and auditable payload tracking; AI/R launched AI/Cockpit One (Aug 10); Snowflake shipped Cortex AI Gateway built on its May acquisition of Natoma to enforce identity, policy and audit at the MCP tool-call level; and HarnessRouter open-sourced its harness gateway today. None of these are applications — they are all tollbooths. The pattern says the durable margin in the agent stack is being claimed at the metering layer, not the app layer.

Graph relationships

SUPPORTS
Cursor
Agentic Stack integrates Cursor session history.
Source finding
SUPPORTS
OpenCode
Agentic Stack integrates OpenCode session history.
Source finding
SUPPORTS
Codex
Agentic Stack integrates Codex session history.
Source finding
SUPPORTS
Claude Code
Agentic Stack integrates Claude Code session history.
Source finding

Source trail

arXiv GitHub arXiv 2609.08368 GitHub arXiv 2609.00052 — AgentProv: Auditing Agentic LLM API Providers via Tool-use Policy Probes arXiv blog.jakesaunders.dev / Hacker News Stack Overflow Blog arXiv 2608.17310 Help Net Security (A10, SelectHub; corroborated by Snowflake's engineering blog, GlobeNewswire/AI-R Aug 10, and harnessrouter.ai)Stack Overflow Blog arXiv 2608.06305 — Beyond Top-K: Replacing Black-Box Retrieval with Interpretable Agentic Operations

Graph sources

entity graphfindings textkg edgeskg entities

Fetching from the wire…