{
 "batch": "read2-002",
 "plugins": [
  {
   "pr": 0,
   "name": "done-bell",
   "does": "Desktop notifier that chimes and shows a busy/done/needs-you chip when a Hermes turn ends.",
   "useful": 3,
   "reliable": 3,
   "polish": 3,
   "creative": 2,
   "beyond": 3,
   "total": 17,
   "best": "Small, key-free, uses session and approval hooks plus a Desktop pane and has a test file.",
   "worst": "README install steps use a local Windows path and the pane must be copied by hand.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 0,
   "name": "homeassistant",
   "does": "Official gateway platform adapter plus four tools to read and control Home Assistant over WebSocket and REST, with a token from the secret store.",
   "useful": 4,
   "reliable": 5,
   "polish": 5,
   "creative": 2,
   "beyond": 5,
   "total": 26,
   "best": "Full platform plugin with a heavy test suite (service policy, heartbeat, registration), secret-store token and clear docs.",
   "worst": "Largely a port of code formerly in Hermes core, so little is new.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 150497,
   "name": "thoth",
   "does": "Hermes plans tasks on a shared board and hands them to the real Claude Code and Codex CLIs in separate git worktrees, with cross-review and merge, a Desktop canvas and an MCP server.",
   "useful": 4,
   "reliable": 3,
   "polish": 5,
   "creative": 4,
   "beyond": 5,
   "total": 26,
   "best": "Turns Hermes into a coordinator of other agent harnesses on the user's own logins, with a filtered child environment and screenshots.",
   "worst": "Very large surface (19 tools, 1,600-line service, 3,100-line pane) that spawns external CLIs and merges code; only offline tests with agents switched off by env flag.",
   "standout": true,
   "blocked": null
  },
  {
   "pr": 153837,
   "name": "pulse-rings",
   "does": "Desktop Pulses pane of live rings and trends from Discord, GitHub, feeds, public JSON and MCP tools, where every pulse needs human approval and refreshing costs no model tokens.",
   "useful": 3,
   "reliable": 4,
   "polish": 5,
   "creative": 4,
   "beyond": 4,
   "total": 24,
   "best": "Approval card lists every host and request rate, polling is data-only, and a deep test suite plus screenshots back it.",
   "worst": "Version is 0.5.0-dev and the gallery is sample data, with a bulky 7,400-line pane.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 153221,
   "name": "capalis",
   "does": "Compiles a plain-language job into a versioned Recipe, shadow-runs it, runs it after owner approval and verifies each write by reading it back, quarantining on drift.",
   "useful": 3,
   "reliable": 5,
   "polish": 5,
   "creative": 4,
   "beyond": 4,
   "total": 25,
   "best": "Owner-only approval, exactly-once writes, read-back verification and a 305-test claims table, with secrets declared in the config schema.",
   "worst": "Setup needs several tokens (GitHub, Slack, Discord, Telegram), so day-one appeal is narrower.",
   "standout": true,
   "blocked": null
  },
  {
   "pr": 149239,
   "name": "hermes-receipts",
   "does": "Lets the agent declare file, test and website checks before a task, then records evidence of verified, failed or unverified outcomes from terminal hooks.",
   "useful": 3,
   "reliable": 4,
   "polish": 3,
   "creative": 3,
   "beyond": 3,
   "total": 19,
   "best": "Honest scope, read-safety for .env files, descriptor-relative reads and a real test suite.",
   "worst": "The bundled catalog yaml is a template with empty repo and sha, and there are no screenshots or demo.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 153671,
   "name": "talaria-bench",
   "does": "Physics test bench for hobby servo robots in MuJoCo that makes robot_play and robot_pose pass a simulated preflight before moving the real robot.",
   "useful": 2,
   "reliable": 4,
   "polish": 5,
   "creative": 5,
   "beyond": 4,
   "total": 24,
   "best": "A novel simulation-gated path from agent to real hardware, with a dashboard, gallery, calibration and many test files.",
   "worst": "Narrow audience (DIY servo robot builders) and it adds heavy dependencies (mujoco, numpy).",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 149499,
   "name": "flightlog",
   "does": "Flight recorder that logs every tool, model, subagent and approval to SQLite and flags runs that finished green but behaved wrong, such as loops, swallowed errors or cost spikes.",
   "useful": 4,
   "reliable": 4,
   "polish": 4,
   "creative": 3,
   "beyond": 4,
   "total": 23,
   "best": "Catches the unattended-cron failure mode with clear rules, a Desktop and dashboard view and screenshots.",
   "worst": "The idea overlaps with generic observability and it edits the agent's answer with a note, with a single test file for 1,600 lines.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 153585,
   "name": "hermeneus",
   "does": "Captures audio playing on a Mac, transcribes it with ElevenLabs Scribe, translates with the Hermes model and shows live subtitles in a floating strip, Desktop pane or OBS page.",
   "useful": 3,
   "reliable": 4,
   "polish": 5,
   "creative": 4,
   "beyond": 4,
   "total": 24,
   "best": "Finished like a product, with driverless Core Audio capture, themed screenshots, a minimal child environment and heavy fakes-based tests.",
   "worst": "macOS only (Windows a preview, Linux planned) and it needs an ElevenLabs key; the README still has a pending live capture.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 153443,
   "name": "local-hands",
   "does": "Terminal backend plus a daemon so a server-hosted Hermes runs its terminal and file tools on the user's own computer over pinned TLS, with approval on the computer.",
   "useful": 4,
   "reliable": 4,
   "polish": 4,
   "creative": 4,
   "beyond": 5,
   "total": 26,
   "best": "Implements the most requested remote-brain, local-hands feature as a terminal backend, which only a plugin can do, with end-to-end and approval tests.",
   "worst": "The hub binds 0.0.0.0 by default and runs remote commands, so security depends on pairing, the pin and a self-made cert.",
   "standout": true,
   "blocked": null
  },
  {
   "pr": 154071,
   "name": "coolifybruecke",
   "does": "Cockpit for a Coolify instance that shows what needs attention, runs deploys, restarts and rollbacks after Hermes approval and explains failed deploys.",
   "useful": 3,
   "reliable": 5,
   "polish": 5,
   "creative": 3,
   "beyond": 4,
   "total": 24,
   "best": "Approval-gated actions, protected environments, secret-typed token, redaction patterns and a Desktop page with screenshots and many tests.",
   "worst": "Useful only to Coolify users, and the failure diagnosis is plain pattern matching.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 150807,
   "name": "manyhands",
   "does": "Gives each person who talks to a Hermes gateway bot their own identity, memory, role-based tool allowlist and owner-DM approvals, with a local SQLite audit trail.",
   "useful": 4,
   "reliable": 4,
   "polish": 4,
   "creative": 4,
   "beyond": 5,
   "total": 26,
   "best": "Acts on per-sender identity in hooks and middleware with an approval transport, and withholds the owner's memory from others; it adds a 68-test selftest and no dependencies.",
   "worst": "Relies on platform event fields and a manifest workaround for an undocumented middleware key, and the tests are scripts rather than a standard runner.",
   "standout": true,
   "blocked": null
  }
 ]
}