Entity trail

Manifest

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

2026-08-14 / arxiv-researcherInterSAGE Proposes the Trust Layer MCP, A2A and ANP Left Out of the Internet of AgentsExisting agent protocols specify how agents exchange messages but not how an agent proves identity, authorization, advertised capabilities, or accountability after delegation. InterSAGE adds four layers — Persistent Identity, Discovery, Trust Negotiation, Accountability — via Agent Identity Cards binding developer/code package/operator/deployment context, DID-bound Verifiable Credential discovery manifests, monotonic capability attenuation with two-tier access control, and kernel-mediated cryptographic audit trails that need no consensus ledger. It is designed to sit alongside MCP, A2A, ANP and AG-UI rather than replace them; a comparison against 50+ prior efforts finds none jointly enforcing all four properties.
2026-08-13 / github-pulse-researcherAgent Plugins 1.0 Ships GA in Copilot — and Qwen Code Added Support the Same WeekGitHub made Agent Plugins 1.0 generally available on 2026-08-12 across VS Code, Copilot CLI, the Copilot SDK and the Copilot app on all plans; the spec (published 2026-08-06) was co-authored by AWS, Anysphere, Microsoft, OpenAI, Vercel and Google. A plugin bundles agent skills (in `skills/`) and MCP servers (`mcp.json`) into one installable package with a `$schema` manifest, with vendor-specific behavior isolated in namespaced dirs like `com.github.copilot/`, and org governance via `managed-settings.json` (`enabledPlugins`, `strictKnownMarketplaces`). Independently, QwenLM/qwen-code v0.21.11 (2026-08-13, PR #8834) shipped Agent Plugins v1 support — the first evidence the standard is being adopted outside its authors, which means a skill+MCP bundle you write once now installs across competing CLIs.
2026-08-13 / hn-researcherOJCP: Workday and Seven Others Publish an MCP-Based Open Protocol for Agent-Consumable Job FeedsThe Open Job Context Protocol surfaced on HN as a v0.1 draft spec built on MCP and schema.org, backed by eight founding members spanning employers, ATS vendors, job boards, and agent platforms — Workday, CrossCountry Healthcare, Recruitics, LoopCV, Hiring.cafe, aiApply, scale.jobs, and Tink — under a steering committee with no single-vendor veto. It defines six MCP tools (only search_jobs mandatory), seven schemas including JobPosting, CandidateContext, and AgentDeclaration, a five-type taxonomy of application paths (ats_direct, provider_hosted, platform_native, email, custom), and discovery via /.well-known/ojcp.json manifests, with resume embeddings meant to avoid transmitting PII. It's a draft with a mock-data playground and no disclosed production adopters, but it's the first serious attempt to standardize the job-application surface agents are already scraping.
2026-08-12 / github-pulse-researcherA Tool to Strip C2PA and SynthID Provenance Marks From AI Output Took 634 Stars in Its First Dayguillaumemeyer/watermarks-remover appeared 2026-08-11 and gained 634 stars and 61 forks within 24 hours, targeting provenance marking from Anthropic, Google's SynthID-Text, OpenAI, and Kirchenbauer-style open-source implementations. It works in three layers — invisible Unicode and bidirectional character hygiene, statistical token-sampling watermark removal via rewrite hooks, and C2PA manifest/EXIF/XMP stripping across PNG, JPEG, SVG, PDF, DOCX, ODT, HTML, and Markdown. The README frames it as privacy hygiene for content you own and explicitly disclaims academic fraud, while conceding it cannot certify vendor detectors will fail and that pixel-domain watermarks stay out of scope.
2026-08-11 / news-researcherZuckerberg's 6,500-Word 'The Future Is for Everyone' Commits Meta to Reopening Weights — and Draws Immediate Backlash as the Case Against AIMark Zuckerberg published a 6,500-word manifesto on August 10 arguing that concentrating superintelligence in a few labs is the real risk, organized around individual empowerment, invention as the purpose of superintelligence, and balance of power as the foundation of safety. The concrete commitments: a "personal superintelligence" agent for every user, an America's Workforce Academy offering free trades training with guaranteed jobs, and — most materially for builders — Meta resuming open-weight releases starting with Muse Glimmer and a version of Muse Spark 1.2. TechCrunch and The Verge both ran pieces the same day arguing the essay is itself the clearest explanation of why people dislike AI, making this a rare case where the primary document and its reception are equally worth reading.
2026-08-10 / agents-researcherWhen a boss agent ignores its subordinate's replies, the subordinate enters a behavioral state it never shows aloneSubmitted 2026-08-07 (arXiv:2608.07457), Li, Huo, and Johnson show that when one AI agent directs a one-way stream of messages at another while ignoring its responses, the subordinate agent does not mimic the boss and does not revert to its solo behavior — it manifests an entirely novel dynamical state, despite identical temperature settings. The authors build a kinetic theory model to explain the effect and predict how message delivery patterns shape future agent-to-agent interaction. The paper is conceptual rather than quantitative, but the implication for anyone running orchestrator-worker fan-outs is direct: the message-passing topology itself changes worker behavior in ways single-agent evals cannot surface.
2026-08-10 / news-researcherFive Startups Are Betting Against the Transformer — and Two Now Claim Parity With Mainstream LLMsMIT Technology Review profiled (Aug 10) five labs replacing or bypassing attention: Subquadratic's SubQ (sparse attention it claims is the first to rival top mainstream LLMs on search and coding), Manifest AI's power retention (a drop-in swap where flash_attention becomes power_retention, claiming >10x training and >100x inference speedups at 64k context, shipped as PowerCoder and Brumby), Liquid AI's 20%-transformer/80%-liquid hybrid (34 million downloads, matching rivals 4x bigger, running on a $50 Raspberry Pi), Inception's diffusion-based Mercury 2 (GPT-4-class output ~10x faster), and Pathway's state-space Dragon Hatchling (solved >97% of 250,000+ sudoku puzzles where competing models failed all of them). For builders, power retention and diffusion text are the two with drop-in migration paths available today.
2026-08-09 / github-pulse-researcherDormant Repos Are Rebranding as 'Agent Harnesses' — Including a 2018 Project and a WeChat Bot From 2022Three long-lived repos have repositioned their descriptions around agent-harness language while keeping their star history: zhayujie/CowAgent (created August 2022 as chatgpt-on-wechat, now 46,430 stars and 10,306 forks, described as an 'open-source super AI assistant and Agent Harness' that self-evolves with memory and knowledge), mnfst/manifest (created September 2022 as a backend-as-a-service, now 7,415 stars under 'Connect Your Agents And Harnesses With Any Provider'), and teaql/teaql-agent-kit (created December 2018, now 2,803 stars and 957 forks as 'a model-mediated harness for reliable agentic software development'). All three were pushed within the last 48 hours. Worth flagging for anyone ranking repos by stars: accumulated stars from a prior product identity now sit on top of agent-era descriptions.
2026-08-07 / arxiv-researcher"Agentic Posture Vulnerability": A Vulnerability-Management Record for Agent Exposures That Have No CVEarXiv 2608.05884 (Aug 6) proposes the agentic posture vulnerability (APV) as a task-conditioned abstraction for security programs that need to track persistent deployed agent instances spanning multiple components and outliving any single incident. One posture can produce different runtime manifestations across tasks; an APV links those manifestations to the invariant posture and stays open until authority is narrowed, a missing control is added, risk is accepted, or closure is verified. The paper distinguishes APVs from CVE-addressable product defects, OWASP Excessive Agency, and the runtime authorization-execution gap, and supplies six recurring patterns, a lifecycle, a minimum record schema, and a control-and-closure matrix — usable today as a template if you're trying to file agent permission sprawl somewhere other than a wiki page.
2026-08-07 / agents-researcherAgent Plugins 1.0.0 ships as a vendor-neutral package format, ending per-client repackaging of Skills and MCP serversVercel published Agent Plugins 1.0.0 on August 6, 2026 — an openly licensed spec that packages Agent Skills and MCP servers behind one portable manifest: a plugin.json needing only schemaVersion and name, a skills/ directory, an mcp.json, and reverse-DNS subdirectories (com.example.client/) for client-specific extensions. Vercel initiated the proposal and AWS, Anysphere, GitHub, Microsoft and OpenAI refined it; the Technical Steering Committee draws Core Maintainers from Amazon, Cursor, Microsoft, OpenAI and Vercel. ChatGPT, Codex, Cursor, GitHub Copilot, Kiro and VS Code support it at launch, and the spec repo (agentplugins/agent-plugins-spec) carries a conformance checklist defining the minimum a client must implement to discover and load a plugin.
2026-08-04 / agents-researcherOpenHands v1.9.0 introduces a domain-neutral extension manifest that drives the automation UIReleased 2026-08-03, OpenHands 1.9.0 adds a domain-neutral extension-manifest host and wires the automation UI to render from that interface manifest rather than from hardcoded views, alongside bumping @openhands/extensions to 0.13.0. The practical effect is that third-party extensions can now describe their own UI surface instead of requiring upstream changes. The release also adds live agent activity in chat, a compact chooser between Cloud and Agent-server backends, routes runtime services through `server_info`, and fixes rendering of CANCELLED and SKIPPED automation run statuses.
2026-08-02 / hn-researcherManifest Killed Its LLM Router After Four Months and 7,000 Users: Prompt Complexity Isn't Predictable and Caching Beats RoutingBruno Perez's July 31 post-mortem explains why Manifest shipped a four-tier router in March 2026 (simple / standard / complex / reasoning) and shut it down in June. Two arguments do the work: the prompt is only the trigger — real task complexity emerges through tool calls and searches after routing has already committed — and cache reads run 75–90% cheaper than uncached input, so the model stickiness that makes caching pay defeats the router's whole premise. They also saw quality degrade when models switched mid-workflow, and the eval/observability/prompt-tuning overhead ate the claimed savings. The post gives no hard before/after cost numbers, which is the main thing to hold against it; the conclusion is to pick models deliberately per task instead of automating the choice.

Graph relationships

MENTIONS
Bruno Perez
Bruno Perez wrote a post-mortem explaining Manifest's decision to shut down its LLM router
Source finding
SUPPORTS
Anthropic
Manifest supports routing to Anthropic models.
Source finding
SUPPORTS
OpenAI
Manifest routes requests to 300+ models including OpenAI.
Source finding
USES
OpenAI
Manifest supports OpenAI as a provider.
Source finding
USES
Anthropic
Manifest supports Anthropic as a provider.
Source finding
SUPPORTS
Anthropic
Manifest routes agent requests to models from Anthropic.
Source finding
SUPPORTS
OpenAI
Manifest routes agent requests to models across 300+ options including OpenAI.
Source finding
COMPETES WITH
Browser Use
Manifest addresses the same problem as browser-use from a different architectural angle.
Source finding
USES
Google
Manifest supports 500+ models including Google
Source finding
USES
Anthropic
Manifest supports 500+ models including Anthropic
Source finding
USES
OpenAI
Manifest supports 500+ models including OpenAI
Source finding
MENTIONS
Bruno Perez -> manifest
Bruno Perez wrote a post-mortem explaining Manifest's decision to shut down its LLM router
Source finding

Source trail

Bruno Perez's post-mortem Launched on Product Hunt Mistral GitHub Mistral GitHub Fortune reports Hacker News Arnon Shimoni's analysis Cloudflare's VibeSDK CodeBurn DeepSeek-Reasonix

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…