{
 "batch": "read1-053",
 "plugins": [
  {
   "pr": 0,
   "name": "done-bell",
   "does": "OS notification, status chip and optional chime that fire when a Hermes turn finishes or approval is needed.",
   "useful": 3,
   "reliable": 4,
   "polish": 4,
   "creative": 2,
   "beyond": 4,
   "best": "Clean, small, well-tested (bell_store, bell_hooks, bell_context all covered) and touches no secrets or user files.",
   "worst": "It is a narrow convenience feature (a notification chime) with little room for depth or differentiation.",
   "standout": false,
   "blocked": null,
   "total": 21
  },
  {
   "pr": 0,
   "name": "homeassistant",
   "does": "Official Nous Research platform plugin that bridges Hermes to Home Assistant: a gateway platform forwarding state-change events and a 4-tool toolset to query/control devices over HA's REST API.",
   "useful": 5,
   "reliable": 5,
   "polish": 5,
   "creative": 3,
   "beyond": 5,
   "best": "Backed by ~2,500 lines of tests (adapter, tools, websocket heartbeat, service policy, plugin registration) and reads its token only via Hermes's env/secret mechanism.",
   "worst": "It is a maintained port of a feature that used to ship inside Hermes core, so it is more a mature integration than a creative new idea.",
   "standout": false,
   "blocked": null,
   "total": 28
  },
  {
   "pr": 153869,
   "name": "agssh",
   "does": "Registers SSH hosts as first-class Hermes tools (run command, read lines, grep, list hosts) over one persistent multiplexed connection per host, returning structured JSON instead of raw shell strings.",
   "useful": 4,
   "reliable": 4,
   "polish": 4,
   "creative": 3,
   "beyond": 4,
   "best": "44 test functions across unit/integration/acceptance suites, a pre_tool_call guard that blocks raw ssh calls to registered hosts, and a documented private-API disclosure (docs/PUBLIC-SURFACE.md).",
   "worst": "Dangerous commands route through Hermes's own approval classifier rather than the plugin forcing approval on every remote command, so safety depends on that classifier being well tuned.",
   "standout": false,
   "blocked": null,
   "total": 23
  },
  {
   "pr": 149744,
   "name": "wellness-nourish",
   "does": "An MCP-server nutrition tracker exposed to Hermes: searches USDA/Open Food Facts/TACO food databases, estimates meals from photos or text, and logs meals, water and goals to local JSON/JSONL files.",
   "useful": 3,
   "reliable": 4,
   "polish": 5,
   "creative": 3,
   "beyond": 4,
   "best": "36 real regression scripts covering concurrency locking, path-traversal defenses and localization, plus an optional API key that is explicitly resolved and forwarded by Hermes rather than read from disk.",
   "worst": "Destructive tools (delete/clear/undo of logged intake) rely on the skill's advisory text for user confirmation rather than a code-enforced approval gate.",
   "standout": false,
   "blocked": null,
   "total": 23
  },
  {
   "pr": 150447,
   "name": "leonardo",
   "does": "Adds Leonardo.AI as a text-to-video and image-to-video generation backend for Hermes's video_gen provider system, polling Leonardo's REST API to completion.",
   "useful": 3,
   "reliable": 3,
   "polish": 3,
   "creative": 3,
   "beyond": 4,
   "best": "Reads its API key correctly through agent.secret_scope rather than the filesystem or raw environment, and is transparent that it does not meter the user's Leonardo spend.",
   "worst": "No test suite at all for a plugin whose only files are __init__.py, plugin.yaml and README.md.",
   "standout": false,
   "blocked": null,
   "total": 20
  },
  {
   "pr": 152850,
   "name": "ussd-platform",
   "does": "A Hermes gateway platform adapter that lets users talk to their agent from any basic phone over USSD (Africa's Talking webhook), paging replies and rendering approvals as numbered keypad menus, with optional SMS relay.",
   "useful": 4,
   "reliable": 4,
   "polish": 5,
   "creative": 4,
   "beyond": 5,
   "best": "Genuinely novel reach for an agent platform (any GSM phone, no app or data plan), with credentials read only through Hermes's scoped secret store, approvals bound to the specific request, and a bundled HTML simulator plus screenshots.",
   "worst": "Its core audience (feature-phone USSD users) is a narrow slice of the installed Hermes base even though the idea itself is compelling.",
   "standout": true,
   "blocked": null,
   "total": 27
  },
  {
   "pr": 153977,
   "name": "trade-preflight",
   "does": "A pre_tool_call/post_tool_call policy firewall that normalizes two simulated ('paper') equity order tools into a canonical form and blocks any order that violates an operator-edited JSON policy, logging an audit trail to local SQLite.",
   "useful": 3,
   "reliable": 5,
   "polish": 5,
   "creative": 4,
   "beyond": 4,
   "best": "By design it needs no broker account, API key, or network access at all, and is fail-closed and deterministic, with an ADR, architecture docs and ~780 lines of edge-case tests.",
   "worst": "Because it only simulates paper trades against two mock adapters, it never actually touches a real brokerage, so its real-world usefulness is mostly as a policy-firewall pattern/demo rather than a deployable trading safeguard.",
   "standout": false,
   "blocked": null,
   "total": 25
  },
  {
   "pr": 152981,
   "name": "claude-pacer",
   "does": "Reads a user's Claude Pro/Max 5-hour and weekly usage windows via Hermes's own usage-fetch API and computes a pacing mode to throttle agents and gate cron jobs so the weekly allowance lasts and some is reserved for the human.",
   "useful": 4,
   "reliable": 5,
   "polish": 4,
   "creative": 3,
   "beyond": 4,
   "best": "About 105 stdlib unit tests covering edge cases like NaN/Infinity rejection and fail-open behavior, and it explicitly never reads, holds or writes a credential itself.",
   "worst": "No Desktop UI or visual component; it is a CLI/tool/slash-command utility whose value depends entirely on users who run heavy scheduled automation against subscription limits.",
   "standout": false,
   "blocked": null,
   "total": 24
  },
  {
   "pr": 151536,
   "name": "card-grader",
   "does": "Computer-vision plugin that measures a trading card's centering from a photo to derive a PSA centering ceiling, checks whitening/clarity, optionally asks the user's vision LLM for a surface opinion, and pulls sold prices and live listings for a grade-or-not verdict.",
   "useful": 3,
   "reliable": 4,
   "polish": 4,
   "creative": 4,
   "beyond": 4,
   "best": "Unusually honest engineering: BENCH.md and bench_odds.json show the author abandoned an initial 'predicted grade' claim after benchmarking against 64 real slabs, downgrading the claim rather than overselling it.",
   "worst": "Its audience is a specific hobby (trading-card grading), so most Hermes users would have no use for it day one.",
   "standout": false,
   "blocked": null,
   "total": 23
  },
  {
   "pr": 153809,
   "name": "living-spaces",
   "does": "A native Hermes Desktop plugin giving bots a persistent, sandboxed canvas ('Spaces') to build dashboards, trackers and mini-apps in chat, backed by local SQLite storage, revisions, and native-cron scheduled upkeep.",
   "useful": 4,
   "reliable": 5,
   "polish": 5,
   "creative": 4,
   "beyond": 5,
   "best": "An unusually large and adversarial test suite (~50 Python/JS test files plus a dedicated harness/ with hostile-webrtc and hostile-nav fixtures) and a file_refs deny-list that explicitly blocks importing .env/credentials.json/*.key files.",
   "worst": "Its surface area (15 tools, a JS sandbox, a desktop bridge, native cron) is large enough that correctness depends on a lot of moving parts staying in sync, which is harder to fully verify by reading than a smaller plugin.",
   "standout": true,
   "blocked": null,
   "total": 28
  },
  {
   "pr": 153467,
   "name": "fixmyscan",
   "does": "A manifest/connector pointing Hermes at an externally hosted MCP server and REST API (mcp.fixmyscan.com) that OCRs and cleans scanned book PDFs/DjVu files for e-readers; no implementation code exists in this repo.",
   "useful": 3,
   "reliable": 2,
   "polish": 3,
   "creative": 2,
   "beyond": 2,
   "best": "Clear, professional manifests (openapi.json, mcp.json, SKILL.md) describing a real external service with its own useful function.",
   "worst": "There is no local implementation to judge at all (only manifests and a SKILL.md) and the plugin itself marks the remote server as 'untrusted', with no test suite and no code a skill plus a remote MCP tool call couldn't also achieve.",
   "standout": false,
   "blocked": null,
   "total": 14
  },
  {
   "pr": 154071,
   "name": "coolifybruecke",
   "does": "A cockpit plugin for a self-hosted Coolify PaaS instance: reports what needs attention across the fleet, shows deployment history, diagnoses failed deploys from logs, and gates deploy/restart/stop/rollback actions behind human approval, with a Desktop dashboard pane.",
   "useful": 4,
   "reliable": 5,
   "polish": 5,
   "creative": 4,
   "beyond": 5,
   "best": "Tokens are declared as secrets and read only via Hermes's secret store; destructive actions require a human-approved one-time ticket and refuse unattended/cron contexts and protected environments by default; a dedicated prompt_injection.log fixture verifies the diagnose tool treats untrusted log text as data, never instructions.",
   "worst": "Its value is tied to a specific self-hosting stack (Coolify), so it is strong but niche rather than broadly applicable to all Hermes users.",
   "standout": true,
   "blocked": null,
   "total": 28
  }
 ]
}