{
  "batch": "read1-052",
  "plugins": [
    {
      "pr": 0,
      "name": "done-bell",
      "does": "Notifies users when Hermes finishes a turn or needs approval through CLI notifications and a Desktop status chip and settings pane.",
      "useful": 4,
      "reliable": 4,
      "polish": 3,
      "creative": 2,
      "beyond": 4,
      "best": "Turn and approval hooks plus Desktop busy-state observation provide a useful capability a prompt cannot supply.",
      "worst": "Setup documentation uses an author's local Windows path and a manual Desktop copy step, and the tests do not exercise Desktop notification delivery.",
      "standout": false,
      "blocked": null,
      "total": 21
    },
    {
      "pr": 0,
      "name": "homeassistant",
      "does": "Connects Hermes to Home Assistant through four device tools, a state-change gateway platform and cron notification delivery.",
      "useful": 4,
      "reliable": 5,
      "polish": 4,
      "creative": 3,
      "beyond": 5,
      "best": "Profile-scoped secrets, validated service selectors, operator entity restrictions and extensive mocked policy and adapter tests support a substantial platform integration.",
      "worst": "Useful only to Home Assistant owners, and entity restrictions are opt-in while the default token can control everything its Home Assistant user permits.",
      "standout": false,
      "blocked": null,
      "total": 26
    },
    {
      "pr": 153678,
      "name": "consequence-map",
      "does": "Provides a decision-audit library and Hermes hooks for correlating approval requests and outcomes, with a separate prototype authorization guard.",
      "useful": 2,
      "reliable": 1,
      "polish": 2,
      "creative": 4,
      "beyond": 3,
      "best": "The library separates host authorization from agent decisions and includes substantial lifecycle, audit and authorization tests.",
      "worst": "The pinned plugin subdirectory imports a sibling implementation package absent from that install layout, catches registration failure, and supplies no active Guard service; the PR explicitly confirms zero registered hooks for that layout.",
      "standout": false,
      "blocked": null,
      "total": 15
    },
    {
      "pr": 150130,
      "name": "hermes-make-plain",
      "does": "Adds a Desktop composer button that rewrites the current draft through Hermes’s auxiliary model route and replaces it without sending.",
      "useful": 3,
      "reliable": 4,
      "polish": 4,
      "creative": 2,
      "beyond": 2,
      "best": "A focused localized UI includes busy, duplicate-click, unload and concurrent-edit guards, with stub-SDK tests.",
      "worst": "The core feature is a wording-cleanup prompt, and preservation of every concrete detail is requested from the model rather than verified.",
      "standout": false,
      "blocked": null,
      "total": 17
    },
    {
      "pr": 149589,
      "name": "hermes-mark-words",
      "does": "Renders model-authored hidden word directives as persistent Desktop underlines, hover explanations and follow-up composer actions.",
      "useful": 3,
      "reliable": 2,
      "polish": 4,
      "creative": 4,
      "beyond": 4,
      "best": "The in-band directive format supports immediate interactive annotations without an extra model request, with detailed UI regression tests.",
      "worst": "It queries internal transcript selectors such as [data-role=\"assistant\"] and .aui-md and observes the document tree, conflicting with the catalog SDK-only policy; annotations also depend on the replying model spotting its own doubtful terms.",
      "standout": false,
      "blocked": null,
      "total": 21
    },
    {
      "pr": 150310,
      "name": "loredocs",
      "does": "Launches a version-pinned local MCP knowledge vault with document CRUD, version history, full-text search and paid semantic search.",
      "useful": 4,
      "reliable": 2,
      "polish": 3,
      "creative": 2,
      "beyond": 3,
      "best": "The backing service implements persistent SQLite and filesystem document storage, versioning and context retrieval beyond a prompt.",
      "worst": "No test suite is present in the pinned tree, the Hermes README documents long-body tool-call failures, and the runtime licence resolver can read the sibling LoreConvo licence store outside Hermes’s secret routes.",
      "standout": false,
      "blocked": null,
      "total": 17
    },
    {
      "pr": 152378,
      "name": "opportunity-check",
      "does": "Searches public agent-work listings, checks GitHub issue and repository signals, and saves up to five issue snapshots for manual rechecks.",
      "useful": 3,
      "reliable": 4,
      "polish": 3,
      "creative": 2,
      "beyond": 3,
      "best": "Bounded requests, redirect refusal and mocked tests accompany clear separation of advertised rewards from verified facts.",
      "worst": "It checks only the first 30 open pull requests and cannot verify funding or AI eligibility, while malformed upstream JSON shapes can escape its narrow exception handling.",
      "standout": false,
      "blocked": null,
      "total": 18
    },
    {
      "pr": 150346,
      "name": "construct",
      "does": "Builds checked parametric scenes into interactive Desktop 3D views, model and print exports, CAD, kits, films and optional NetLogo visualizations.",
      "useful": 4,
      "reliable": 4,
      "polish": 5,
      "creative": 5,
      "beyond": 5,
      "best": "A bundled readable geometry engine, twelve tools, integrated viewers, provenance and export readback form an unusually complete creative workflow.",
      "worst": "CAD, capture, film and simulation paths depend on separately installed native programs, and the PR reports end-to-end platform testing only on macOS.",
      "standout": true,
      "blocked": null,
      "total": 28
    },
    {
      "pr": 150644,
      "name": "hermes-canvas",
      "does": "Connects an editable Desktop plan canvas to real tool writes, check exit codes and subagent activity, with mid-turn steering and removed-box write vetoes.",
      "useful": 5,
      "reliable": 4,
      "polish": 5,
      "creative": 5,
      "beyond": 5,
      "best": "Seven hooks make the visual plan operational: user edits reach running agents, failed checks hold completion, and persisted history supports replay.",
      "worst": "Shell and execute_code write detection is explicitly best-effort, and the veto allows calls on unexpected hook errors, so removed-box enforcement is incomplete.",
      "standout": true,
      "blocked": null,
      "total": 29
    },
    {
      "pr": 151037,
      "name": "herd",
      "does": "Synchronizes Hermes configuration, skills, memory and SOUL across paired machines through encrypted stores and per-key conflict-aware merging.",
      "useful": 4,
      "reliable": 4,
      "polish": 4,
      "creative": 4,
      "beyond": 5,
      "best": "Machine-specific settings and credential-bearing config sections are preserved locally, with encrypted objects, reversible pulls and substantial conflict and safety tests.",
      "worst": "It requires a reachable shared folder or configured git store, and automatic interactive-session pulls can replace instruction-bearing files from a trusted peer without a fresh sync_now approval.",
      "standout": false,
      "blocked": null,
      "total": 26
    },
    {
      "pr": 151504,
      "name": "hermes-manas-aegis",
      "does": "Applies pattern-based tool-call blocking and argument rewrites, with decoy credential paths, local audit history and an authenticated loopback security dashboard.",
      "useful": 3,
      "reliable": 2,
      "polish": 3,
      "creative": 3,
      "beyond": 4,
      "best": "Real pre-tool enforcement, redacted audit state, simulation tests and a sandboxed Desktop dashboard go beyond advisory prompting.",
      "worst": "The pre-tool path performs synchronous DNS resolution under the core lock without an explicit timeout, and bounded pattern scans and duplicate-call heuristics cannot provide the broad protection implied by its security framing.",
      "standout": false,
      "blocked": null,
      "total": 19
    },
    {
      "pr": 153040,
      "name": "model-forge",
      "does": "Inventories local and configured compute targets, recommends models, downloads weights, plans or applies chat-model selection, and submits media jobs to existing engines.",
      "useful": 4,
      "reliable": 2,
      "polish": 2,
      "creative": 3,
      "beyond": 4,
      "best": "Hardware-aware model selection and real download and media-job paths address useful local-inference setup work, with mocked planner and tool tests.",
      "worst": "The pinned tools package does not deliver the promised automatic engine lifecycle or its separately packaged Desktop/providers, and authenticated downloads use default redirect handling without rechecking the token’s destination host.",
      "standout": false,
      "blocked": null,
      "total": 19
    }
  ]
}