{
  "batch": "read1-040",
  "plugins": [
    {
      "pr": 0,
      "name": "done-bell",
      "does": "Rings and shows status when a Hermes turn ends, with optional sound, desktop pane, and approval-state indication.",
      "useful": 3,
      "reliable": 4,
      "polish": 4,
      "creative": 3,
      "beyond": 4,
      "total": 22,
      "best": "It turns an otherwise easy-to-miss completed or approval-waiting turn into a desktop notification and status signal.",
      "worst": "It is a narrow convenience plugin, and the README still describes a somewhat manual desktop installation path.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 0,
      "name": "homeassistant",
      "does": "Connects Hermes to Home Assistant through a filtered WebSocket gateway and four REST tools for reading and controlling devices.",
      "useful": 5,
      "reliable": 5,
      "polish": 5,
      "creative": 3,
      "beyond": 5,
      "total": 28,
      "best": "It is a complete, official-quality platform integration with tests, reconnect behavior, filtering, cooldowns, and explicit dangerous-service guards.",
      "worst": "Its usefulness depends on operating a Home Assistant instance and its token, so it is not universal despite being broadly valuable.",
      "standout": true,
      "blocked": null
    },
    {
      "pr": 153347,
      "name": "hermes-shadow-clone",
      "does": "Runs concurrent Gemini personas, an arena judge, claim verification, failover, and optional Telegram reporting for second opinions.",
      "useful": 4,
      "reliable": 2,
      "polish": 3,
      "creative": 4,
      "beyond": 5,
      "total": 23,
      "best": "The concurrent adversarial/architect/performance perspectives make model deliberation an actual Hermes tool rather than a prompt convention.",
      "worst": "The pinned snapshot has only a small validation test surface for a networked multi-model system, plus optional Telegram code and a simulated offline path that make operational behavior less convincing.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 153707,
      "name": "neuron-steer",
      "does": "Finds contrastive sparse MLP-neuron circuits in a local model, stores them, and scales or ablates them during generation with dose scans.",
      "useful": 3,
      "reliable": 4,
      "polish": 4,
      "creative": 5,
      "beyond": 5,
      "total": 26,
      "best": "It exposes a genuinely novel model-level capability based on forward-pass neuron attribution, with lazy loading, download gates, and substantial unit tests.",
      "worst": "The audience is limited to users with compatible local Transformer models and substantial compute, and the large dependency/runtime path remains harder to operate than ordinary plugins.",
      "standout": true,
      "blocked": null
    },
    {
      "pr": 153211,
      "name": "spotify-desktop",
      "does": "Adds a Hermes Desktop status-bar Spotify Connect remote with playback, search, playlists, queue, volume, and device transfer controls.",
      "useful": 4,
      "reliable": 3,
      "polish": 4,
      "creative": 3,
      "beyond": 4,
      "total": 22,
      "best": "It delivers a polished Desktop surface while reusing Hermes's existing Spotify authentication and token-refresh machinery.",
      "worst": "The pinned repository contains no tests and depends on private/internal Spotify integration seams, so compatibility failures are easy to imagine.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 153406,
      "name": "agent-data-tools",
      "does": "Provides three MCP tools that discover and call a hosted catalog of hundreds of web, data, finance, chain, compliance, AI, and utility tools.",
      "useful": 5,
      "reliable": 3,
      "polish": 4,
      "creative": 3,
      "beyond": 4,
      "total": 23,
      "best": "The compact find/details/call interface gives Hermes broad live-data reach without flooding context with hundreds of separate tool definitions.",
      "worst": "Nearly all value and reliability are delegated to an external paid service outside the pinned code, with no local implementation or tests to inspect.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 149643,
      "name": "adextract",
      "does": "Connects Hermes to a read-only hosted MCP service for competitor ad-library search and image/video creative analysis across four platforms.",
      "useful": 3,
      "reliable": 3,
      "polish": 4,
      "creative": 3,
      "beyond": 4,
      "total": 21,
      "best": "It packages a focused, practical marketing-research workflow with platform-specific skills and an explicit read-only/cost disclosure.",
      "worst": "The entry is configuration and instructional text only; the core search and analysis service is remote, paid, and not testable from the pinned snapshot.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 153326,
      "name": "sharewise",
      "does": "Audits photo/video metadata and local visual privacy risks, lets the user review masks, and makes verified sanitized copies without changing originals.",
      "useful": 5,
      "reliable": 4,
      "polish": 5,
      "creative": 4,
      "beyond": 5,
      "total": 28,
      "best": "It combines local metadata checks, macOS Vision detection, human review, irreversible-looking redaction, and post-copy verification into a clear privacy workflow.",
      "worst": "The macOS Vision path and several heavyweight imaging dependencies create platform-specific setup risk, while the visual detector cannot guarantee every sensitive feature is found.",
      "standout": true,
      "blocked": null
    },
    {
      "pr": 153854,
      "name": "session-cassettes",
      "does": "Records read-only session results and replays the same request against another model or prompt from a workspace snapshot, refusing live tool calls.",
      "useful": 4,
      "reliable": 4,
      "polish": 5,
      "creative": 5,
      "beyond": 5,
      "total": 28,
      "best": "Its middleware-level claimed replay, pins, explicit PASS/FAIL/DIVERGED/INVALID verdicts, and extensive security/test documentation make model comparison reproducible and safer.",
      "worst": "The pinned code documents major untested platform limits, and replay can become INVALID or unsafe if the claim race or gateway conditions are not met.",
      "standout": true,
      "blocked": null
    },
    {
      "pr": 153876,
      "name": "docket",
      "does": "Runs a trades-business workflow from voice or text job intake through deterministic price-book quotes, approval, customer acceptance, booking, follow-ups, and invoicing.",
      "useful": 5,
      "reliable": 4,
      "polish": 5,
      "creative": 4,
      "beyond": 5,
      "total": 28,
      "best": "It is an unusually complete product with Desktop, chat, customer web pages, PDFs, pricing tests, explicit approval before sending, and protections against unpriced extra work.",
      "worst": "It has many stateful surfaces, email/web-server/payment-adjacent integrations, and a very broad codebase, so deployment and edge-case reliability remain substantial risks.",
      "standout": true,
      "blocked": null
    },
    {
      "pr": 149514,
      "name": "lemma",
      "does": "Extracts scientific claims, runs bounded numerical audits with positive controls in a sandbox, and preserves evidence-backed verdict artifacts and traces.",
      "useful": 4,
      "reliable": 4,
      "polish": 5,
      "creative": 5,
      "beyond": 5,
      "total": 28,
      "best": "The evidence trail, mandatory control, sandbox boundary, provider credential route, and Desktop audit room turn paper checking into a real agent capability.",
      "worst": "It ships a large dependency stack and relies on LLM-generated audit scripts and external paper/providers, leaving significant operational complexity despite the safety defaults.",
      "standout": true,
      "blocked": null
    },
    {
      "pr": 151804,
      "name": "dress-rehearsal",
      "does": "Intercepts Hermes side effects into a reviewable numbered script, lets the user strike steps, and replays exactly the approved calls with drift checks.",
      "useful": 5,
      "reliable": 5,
      "polish": 5,
      "creative": 4,
      "beyond": 5,
      "total": 29,
      "best": "The pre-tool hook plus exact replay, diffs, risk flags, approval selection, and stale-file refusal directly addresses the gap between an agent's plan and its actual actions.",
      "worst": "Its conservative shell classifier and replay assumptions can reject benign workflows or leave untested platform-specific behavior, even though the core fail-closed design is strong.",
      "standout": true,
      "blocked": null
    }
  ]
}