{
  "batch": "read2-010",
  "plugins": [
    {
      "pr": 0,
      "name": "done-bell",
      "does": "Adds turn-completion and approval-needed notifications, a status tool, and a small Desktop status control.",
      "useful": 3,
      "reliable": 4,
      "polish": 4,
      "creative": 2,
      "beyond": 4,
      "total": 21,
      "best": "It uses Hermes lifecycle and approval hooks to make an agent's otherwise invisible state visible away from the chat window.",
      "worst": "It is a comparatively small notification feature, and the Desktop installation path includes a manual copy workaround.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 0,
      "name": "homeassistant",
      "does": "Connects Hermes to Home Assistant through a WebSocket gateway adapter and guarded REST tools for listing, reading, and controlling entities.",
      "useful": 5,
      "reliable": 5,
      "polish": 5,
      "creative": 3,
      "beyond": 5,
      "total": 28,
      "best": "It is a mature platform integration with extensive tests, reconnection behavior, scoped secrets, and explicit service/entity safety controls.",
      "worst": "Its value and setup depend on the user already operating a Home Assistant instance and granting it a token.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 153481,
      "name": "hermes-wake",
      "does": "Persists watches for detached systemd units, processes, or files and wakes the originating messaging chat when the work completes.",
      "useful": 4,
      "reliable": 5,
      "polish": 4,
      "creative": 4,
      "beyond": 5,
      "total": 27,
      "best": "It closes a genuine agent-loop gap with durable SQLite state, compare-and-set completion, retry handling, and careful chat scoping.",
      "worst": "Delivery relies on a running Hermes gateway and Kanban integration, so an offline gateway delays the wake.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 153826,
      "name": "googlehome",
      "does": "Turns a Google Home or Nest Cast speaker into a Hermes voice endpoint with host-microphone input, TTS, Cast playback, tools, and a Desktop control surface.",
      "useful": 4,
      "reliable": 5,
      "polish": 5,
      "creative": 5,
      "beyond": 5,
      "total": 29,
      "best": "A substantial stdlib Cast platform integration combines routed profiles, TTS/STT, authenticated loopback control, Desktop UI, and deep fake-device tests.",
      "worst": "It needs compatible LAN hardware plus host microphone and speech configuration, which makes it less immediately useful than a general-purpose plugin.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 153844,
      "name": "fine-print",
      "does": "Observes Hermes hook data to identify silent context and tool-result limits, then records them and surfaces actionable notices and model notes.",
      "useful": 4,
      "reliable": 4,
      "polish": 5,
      "creative": 5,
      "beyond": 5,
      "total": 28,
      "best": "It gives agents and users visibility into core context loss through a carefully bounded, heavily fixture-tested observability layer.",
      "worst": "Its detectors intentionally target exact Hermes 0.23 formats, so unrecognized future core formats become silent rather than being inferred.",
      "standout": true,
      "blocked": null
    },
    {
      "pr": 151704,
      "name": "guild",
      "does": "Creates a Git-backed, reviewer-approved shared skill library that agents can propose to and sync while preventing direct edits to team skills.",
      "useful": 4,
      "reliable": 4,
      "polish": 5,
      "creative": 4,
      "beyond": 5,
      "total": 27,
      "best": "It treats learned skills as collaborative, reviewable artifacts with scrubbing, provenance, human-only approval, and end-to-end multi-member tests.",
      "worst": "Its local pre-tool guard cannot inspect arbitrary code run through execute_code, and remote branch protection remains the team's responsibility.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 153294,
      "name": "ci-red-to-green",
      "does": "Watches GitHub Actions failures and offers a user-approved source-only fix PR only after recorded local regression and full-suite proof gates pass.",
      "useful": 5,
      "reliable": 5,
      "polish": 5,
      "creative": 5,
      "beyond": 5,
      "total": 30,
      "best": "It converts an agent coding loop into a rigorously gated product workflow, with explicit approval, proof ledger, protected-file checks, demo, Desktop UI, and tests.",
      "worst": "It is deliberately limited to GitHub Actions projects using pytest or go test and cannot prove fixes requiring CI-only services or secrets.",
      "standout": true,
      "blocked": null
    },
    {
      "pr": 152481,
      "name": "argus",
      "does": "Builds an intent-level view of Hermes history and skill changes, uses a configured model to derive review proposals, and requires human decisions before applying them.",
      "useful": 4,
      "reliable": 4,
      "polish": 5,
      "creative": 5,
      "beyond": 5,
      "total": 28,
      "best": "It is a carefully engineered governance and learning-feedback system with profile-aware history, explicit apply checks, dashboard support, and broad tests.",
      "worst": "Derivation can be expensive and transmits selected local history to the configured model; native Windows is unsupported.",
      "standout": true,
      "blocked": null
    },
    {
      "pr": 151245,
      "name": "freeze",
      "does": "Records a completed task, compiles its dataflow into a checked workflow, and replays it through Hermes tools without model calls until structural drift is detected.",
      "useful": 4,
      "reliable": 4,
      "polish": 4,
      "creative": 5,
      "beyond": 5,
      "total": 27,
      "best": "It makes repeatable agent work materially cheaper while preserving approvals, drift detection, output-write protection, and a constrained subprocess for compiled calculations.",
      "worst": "It cannot detect semantic drift that preserves a data shape, and its generated-code isolation is explicitly defense in depth rather than a hard sandbox.",
      "standout": true,
      "blocked": null
    },
    {
      "pr": 154067,
      "name": "hermes-design-studio",
      "does": "Provides a revision-checked Hermes Desktop image workshop with effect stacks, human-approved GLSL drafts, recipes, previews, and PNG export.",
      "useful": 3,
      "reliable": 4,
      "polish": 5,
      "creative": 4,
      "beyond": 5,
      "total": 26,
      "best": "It is an unusually complete Desktop application with real image tooling, concurrency checks, honest renderer status, security limits, documentation, and extensive recorded verification.",
      "worst": "It requires an enabled WebGL2 Desktop renderer on a POSIX gateway and its GLSL checks do not claim to prevent GPU hangs.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 149159,
      "name": "understudy",
      "does": "Auditions candidate models against sampled local Hermes decisions, judges blind next-step replays, measures uncertainty and cost, and presents reports in chat, CLI, and Desktop.",
      "useful": 4,
      "reliable": 5,
      "polish": 5,
      "creative": 4,
      "beyond": 5,
      "total": 28,
      "best": "It lets a user's own agent evaluate model substitutions with redaction, capability consent, deterministic sampling, budget controls, baseline noise measurement, and strong test coverage.",
      "worst": "It evaluates individual next steps rather than full task completion, and its model judge necessarily adds uncertainty.",
      "standout": true,
      "blocked": null
    },
    {
      "pr": 153978,
      "name": "imprint",
      "does": "Adds a Hermes Desktop source notebook that produces cited, brand-aware decks, one-pagers, graphics, and audio briefs with export and source checks.",
      "useful": 4,
      "reliable": 4,
      "polish": 5,
      "creative": 5,
      "beyond": 5,
      "total": 28,
      "best": "This is a deeply polished source-to-deliverable Desktop product with citation-aware outputs, number and quote checks, offline export code, safety controls, and a very large test suite.",
      "worst": "Its breadth brings significant runtime dependencies and optional browser/model setup, and generated designs still require human review for factual framing.",
      "standout": true,
      "blocked": null
    }
  ]
}