{
 "batch": "read1-090",
 "plugins": [
  {
   "pr": 0,
   "name": "done-bell",
   "does": "Rings an OS notification and optional chime when a Hermes turn finishes, with a busy/done/needs-you status chip in Desktop.",
   "useful": 3,
   "reliable": 3,
   "polish": 3,
   "creative": 2,
   "beyond": 3,
   "total": 17,
   "best": "Small, honest scope: no keys, hooks on session end and approval, and approval commands are deliberately not stored.",
   "worst": "Thin tests (9) and a README with author-specific Windows paths and a manual pane-copy workaround.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 0,
   "name": "homeassistant",
   "does": "Official gateway platform adapter plus four tools that read and control Home Assistant devices over WebSocket and REST.",
   "useful": 4,
   "reliable": 5,
   "polish": 4,
   "creative": 3,
   "beyond": 5,
   "total": 26,
   "best": "A full platform plugin with entity allow/deny lists, cooldowns, a token declared as a password env var and about 176 tests.",
   "worst": "Mostly a port of what used to be in Hermes core, so little is new.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 153984,
   "name": "evident-evidence",
   "does": "Opens a site in a locked-down headless Chromium, has the configured model review how it comes across to buyers, and writes an offline HTML report with annotated screenshots and re-runnable checks.",
   "useful": 3,
   "reliable": 3,
   "polish": 3,
   "creative": 4,
   "beyond": 3,
   "total": 19,
   "best": "Serious safety design: origins denied by default, DNS-pinned transport, delete gated on confirm_delete, model access only through ctx.llm, and 72 unit tests plus integration scripts.",
   "worst": "Very dense one-line code (report.py about 100KB), licensing 'not yet specified', a system-prompt section that claims routing 'even when the previous topic was another plugin', and report CTA links to the author's commercial service.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 153935,
   "name": "clearing",
   "does": "The Clearing: a Desktop writing editor in which the user's own Hermes model returns notes anchored to exact quotes, the writer accepts each by click, and a built-in check and WordNet thesaurus work without a model.",
   "useful": 3,
   "reliable": 4,
   "polish": 5,
   "creative": 4,
   "beyond": 4,
   "total": 24,
   "best": "Product-grade Desktop page with readable source in desktop/src, redaction forced before model calls, no keys, about 216 Python and 110 JS tests, and many screenshots.",
   "worst": "Niche audience (prose writers); the 670KB built bundle includes a UI kit not in the repo, it is tested on macOS only, and STATUS.md is partly stale.",
   "standout": true,
   "blocked": null
  },
  {
   "pr": 152706,
   "name": "circuit-bench",
   "does": "An ESP32 workbench that checks pins against Espressif data, calculates parts, draws schematics, and builds, flashes, monitors and decodes crashes, with a Desktop bench and a simulated board.",
   "useful": 3,
   "reliable": 5,
   "polish": 5,
   "creative": 5,
   "beyond": 4,
   "total": 26,
   "best": "Hardware actions go through a pre_tool_call approval keyed to port, MAC and firmware SHA; allow_flash is off by default, it fails closed, and roughly 2,900 tests cover it.",
   "worst": "Only ESP32 hobbyists will use it, real boards are tested on Linux only, and a /yolo session bypasses the approval prompts (disclosed).",
   "standout": true,
   "blocked": null
  },
  {
   "pr": 153083,
   "name": "signet-drive",
   "does": "Gives the agent its own encrypted Signet Drive (save, fetch, list, restore through the signet CLI) and backs up MEMORY.md and USER.md automatically after memory writes.",
   "useful": 3,
   "reliable": 3,
   "polish": 4,
   "creative": 4,
   "beyond": 4,
   "total": 22,
   "best": "Memory backup through post_tool_call and session hooks, a restore that saves the current state first, no credential held by the plugin, and 73 tests against a fake client.",
   "worst": "signet_drive_save will upload any local file the agent names, with no path allow-list; the 'read-only' /files route can create a share and blocks the event loop; macOS only.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 151009,
   "name": "icon-index",
   "does": "Bundles a catalog of 40 CC0 3D icons that the agent can search, embed or save into a project, plus a Desktop gallery pane.",
   "useful": 2,
   "reliable": 4,
   "polish": 4,
   "creative": 3,
   "beyond": 3,
   "total": 19,
   "best": "The built desktop/plugin.js matches its readable source, and icon_save never overwrites, with a size cap and timeout; 12 tests.",
   "worst": "Narrow use (40 icons from the author's own library), and the 3D viewer and recolouring run on a remote page that is not in the repo.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 150211,
   "name": "loci",
   "does": "Manifest, MCP pointer and two skills connecting Hermes to the closed-source Loci Mac app's local MCP server for summaries, extraction, search and transcription with local models.",
   "useful": 3,
   "reliable": 2,
   "polish": 3,
   "creative": 3,
   "beyond": 3,
   "total": 17,
   "best": "Careful token handling: Hermes's server_json discovery attaches the token, the server is marked untrusted, and model pulls need authorization.",
   "worst": "No executable code at all; every tool lives in a closed app needing macOS 26.2+ on Apple silicon, and the only test is a manifest validation.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 150946,
   "name": "shelldrop",
   "does": "A skill that has the agent set up SSH, tmux, Tailscale and keep-awake on a Mac or Linux box for an iPhone SSH app, plus a remote MCP server that returns fixed help text.",
   "useful": 2,
   "reliable": 2,
   "polish": 3,
   "creative": 2,
   "beyond": 2,
   "total": 13,
   "best": "A small, stateless, readable MCP worker with vitest tests and read-only annotations.",
   "worst": "The remote tools repeat the skill's static text, the live server was never checked, and the skill performs sudo, curl|sh and authorized_keys changes guided only by prompt text.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 153305,
   "name": "wiro",
   "does": "A model-provider plugin that adds the Wiro AI OpenAI-compatible gateway, with a table of 14 models and a reasoning-effort mapping.",
   "useful": 3,
   "reliable": 4,
   "polish": 3,
   "creative": 2,
   "beyond": 4,
   "total": 20,
   "best": "WIRO_API_KEY is declared as a secret and handled only by Hermes, and CI runs integration tests against Hermes releases and main.",
   "worst": "The capabilities are a hardcoded snapshot although the docstring says they come from the live catalog, and background tasks are billed from the wallet through a fixed paid model.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 150653,
   "name": "hermes-ego",
   "does": "An MCP pointer, plus a skill, to a token-protected local relay and Mac daemon that let Hermes read and save pages in the user's ego(lite) browser.",
   "useful": 2,
   "reliable": 2,
   "polish": 3,
   "creative": 3,
   "beyond": 3,
   "total": 16,
   "best": "The relay code in the repo root compares tokens in constant time, rejects Origin headers, and blocks private-address image fetches, with about 90 tests.",
   "worst": "The plugin installs none of that code; the README says it 'was not run against a live Hermes process', the token is also kept on disk, and the browser download is unverified.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 150860,
   "name": "falsify",
   "does": "Has the coding agent pre-register a claim and a check, then runs the check on its change and on reverted copies hunk by hunk, marks checks that pass either way VOID, and records results in a hash-chained ledger.",
   "useful": 3,
   "reliable": 4,
   "polish": 3,
   "creative": 5,
   "beyond": 4,
   "total": 23,
   "best": "A real Hermes approval gate (pre_tool_call plus post_approval_response grants), drafter secrets read via secret_scope, and about 79 tests that use real git repos.",
   "worst": "No sandbox (disclosed), agent-chosen timeouts with no cap, git calls without timeouts, and its headline study results rest on recordings not in the repo.",
   "standout": false,
   "blocked": null
  }
 ]
}