<?xml version="1.0" encoding="UTF-8"?>
<feed xmlns="http://www.w3.org/2005/Atom">
  <title>Andrey Neprel — announcements</title>
  <subtitle>Launches and milestones from @neprel.</subtitle>
  <link href="https://neprel.ai/feed.xml" rel="self" type="application/atom+xml"/>
  <link href="https://neprel.ai/" rel="alternate" type="text/html"/>
  <id>https://neprel.ai/</id>
  <updated>2026-08-19T17:39:14Z</updated>
  <author>
    <name>Andrey Neprel</name>
    <uri>https://neprel.ai/</uri>
    <email>ask@neprel.ai</email>
  </author>

  <entry>
    <title>HINT 1.3 — the spec now reaches coding agents over MCP</title>
    <link href="https://github.com/open-hint-dev/hint/releases/tag/v1.3.0" rel="alternate" type="text/html"/>
    <id>https://github.com/open-hint-dev/hint/releases/tag/v1.3.0</id>
    <updated>2026-08-19T17:39:14Z</updated>
    <published>2026-08-19T17:39:14Z</published>
    <summary>HINT ships an MCP server, so Claude Code, Codex, Cursor and VS Code retrieve the governing spec as a tool call rather than a shell command — and one bootstrap prompt configures whichever client the agent is running in.</summary>
  </entry>

  <entry>
    <title>agent2you released — a team of chat-resident AI agents, from declarative manifests</title>
    <link href="https://github.com/neprel/agent2you" rel="alternate" type="text/html"/>
    <id>https://github.com/neprel/agent2you</id>
    <updated>2026-08-19T11:28:58Z</updated>
    <published>2026-08-19T11:28:58Z</published>
    <summary>Extracted from a production fleet and open sourced under MIT. Each agent is one container — chat presence, failover, and a coding agent as the brain, each signed into its own subscription, with no provider API keys anywhere. One prompt stands up a supervisor that then hires the rest of the team.</summary>
  </entry>

  <entry>
    <title>acp2api measured against every agent in the ACP registry — all 38 of them</title>
    <link href="https://github.com/neprel/acp2api/blob/main/docs/agents.md" rel="alternate" type="text/html"/>
    <id>https://github.com/neprel/acp2api/blob/main/docs/agents.md</id>
    <updated>2026-08-14T00:00:00Z</updated>
    <published>2026-08-14T00:00:00Z</published>
    <summary>Compatibility measured rather than claimed: every registry agent installed, initialized and driven through the bridge. Five verified end to end, fifteen behind their own account gate, seven needing a model, nine unable to finish a turn — each with the reason recorded.</summary>
  </entry>

  <entry>
    <title>Two Mattermost MCP servers released — Boards and Playbooks</title>
    <link href="https://github.com/neprel/mattermost-boards-mcp" rel="alternate" type="text/html"/>
    <id>https://github.com/neprel/mattermost-boards-mcp</id>
    <updated>2026-08-13T13:20:45Z</updated>
    <published>2026-08-13T13:20:45Z</published>
    <summary>Mattermost Boards and Playbooks as tools an agent can actually use. Both 1.0 releases were verified end to end against a live Mattermost 11.10.0 as a real bot account — which is how each of them found a bug that a full green test suite had missed. https://github.com/neprel/mattermost-playbooks-mcp</summary>
  </entry>

  <entry>
    <title>acp2api released — any ACP coding agent behind the OpenAI API</title>
    <link href="https://github.com/neprel/acp2api" rel="alternate" type="text/html"/>
    <id>https://github.com/neprel/acp2api</id>
    <updated>2026-08-09T19:20:32Z</updated>
    <published>2026-08-09T19:20:32Z</published>
    <summary>Use Claude Code, Codex or any ACP agent from anything that speaks the OpenAI API. Each agent you configure becomes a model id, and the turn spends the subscription the CLI is already signed into — no API key is read, replayed or forwarded.</summary>
  </entry>

  <entry>
    <title>HINT 1.0 released — Spec-as-Source for any repository</title>
    <link href="https://openhint.dev/" rel="alternate" type="text/html"/>
    <id>https://openhint.dev/</id>
    <updated>2026-06-11T14:33:54Z</updated>
    <published>2026-06-11T14:33:54Z</published>
    <summary>The specification becomes the artifact you maintain: intent lives in Markdown-native .hint files next to what they govern, retrieval returns the part of the spec that applies to a path, and drift is reported while the correction is still cheap. No model in the loop.</summary>
  </entry>
</feed>
