{
  "batch": "read1-100",
  "plugins": [
    {
      "pr": 0,
      "name": "done-bell",
      "does": "Notifies the user when Hermes finishes a turn or awaits approval, with a chime, Desktop status chip and mute controls.",
      "useful": 4,
      "reliable": 3,
      "polish": 3,
      "creative": 2,
      "beyond": 4,
      "total": 20,
      "best": "Actual lifecycle hooks and a Desktop busy-state watcher make completion and approval notifications automatic.",
      "worst": "The README requires manual Windows Desktop copying, and the small tests cover backend state and hooks rather than real Desktop notifications.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 0,
      "name": "homeassistant",
      "does": "Connects Hermes to Home Assistant device control, filtered state-change events and notification delivery.",
      "useful": 4,
      "reliable": 4,
      "polish": 4,
      "creative": 3,
      "beyond": 5,
      "total": 25,
      "best": "A genuine gateway platform combines event subscriptions, reconnect backoff and scoped smart-home tools with tested entity restrictions.",
      "worst": "Entity allow/deny restrictions are opt-in and the default Home Assistant URL uses plain HTTP, so operators must configure transport and device access carefully.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 154031,
      "name": "infra-atlas",
      "does": "Builds a live infrastructure graph from Markdown runbooks and discovery, gates infrastructure operations on prior reading, and tracks drift and downtime.",
      "useful": 4,
      "reliable": 1,
      "polish": 5,
      "creative": 4,
      "beyond": 5,
      "total": 24,
      "best": "The integrated topology UI, drift reconciliation and read-before-touch hooks directly improve how Hermes operates the user’s own infrastructure.",
      "worst": "Its credentials.py implements a separate file-backed credential vault with plaintext fallback, violating catalog rule 11, and auto-heal defaults to executing local restart commands.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 153202,
      "name": "rekindra",
      "does": "Recovers images and text from local folders and lets Hermes turn the evidence into animated browser scenes with preview and WebM export.",
      "useful": 3,
      "reliable": 2,
      "polish": 4,
      "creative": 4,
      "beyond": 3,
      "total": 19,
      "best": "A substantial local recovery and rendering pipeline separates source evidence from generated interpretation and checks previews before export.",
      "worst": "The pinned preview.py passes os.environ.copy() to Chrome/Edge/Chromium, exposing any inherited provider secrets to the browser child instead of using a minimal environment.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 152176,
      "name": "gmi-sandbox",
      "does": "Runs Hermes terminal and file operations in disposable GMI Cloud sandboxes through a native terminal-environment provider.",
      "useful": 3,
      "reliable": 3,
      "polish": 3,
      "creative": 2,
      "beyond": 5,
      "total": 21,
      "best": "The provider extends where Hermes executes code while retaining command guards, profile-scoped secrets and explicit cancellation and cleanup handling.",
      "worst": "The author explicitly reports offline-only verification, so real provisioning, file operations and billable-resource cleanup remain unproven.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 152578,
      "name": "penthera",
      "does": "Packages a skill that guides Hermes through the Penthera CLI for local repository, dependency, URL and agent-configuration security scans.",
      "useful": 3,
      "reliable": 3,
      "polish": 3,
      "creative": 2,
      "beyond": 1,
      "total": 13,
      "best": "The skill provides detailed workflows, pinned CLI commands and approval guidance backed by an extensive standalone scanner repository.",
      "worst": "The pinned plugin folder registers no tools, hooks or MCP server and is entirely a CLI-backed skill, with installation documentation still aimed at Claude Code.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 153526,
      "name": "idea-board",
      "does": "Mines the user’s chats and available Hermes capabilities into an evidence-backed Desktop board of suggested tasks and connectors, learning from dismissals.",
      "useful": 4,
      "reliable": 4,
      "polish": 5,
      "creative": 4,
      "beyond": 4,
      "total": 25,
      "best": "Capability-aware validation, durable dismissal rules and a careful Desktop setup turn agent suggestions into a usable personal workflow without automatically sending them.",
      "worst": "A run can consume 70–80k tokens as disclosed, and capability discovery and duplicate checks depend partly on undocumented Hermes imports with fallbacks.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 153413,
      "name": "vigil",
      "does": "Detects refusal-like replies in model middleware, triggers Hermes’s configured fallback chain and exposes context-bound reviews and observed retry outcomes.",
      "useful": 3,
      "reliable": 3,
      "polish": 4,
      "creative": 2,
      "beyond": 5,
      "total": 22,
      "best": "Model middleware plus a substantial Desktop review UI provide genuine routing behavior and distinguish observed fallback outcomes from mere detections.",
      "worst": "Detection remains heuristic and named profiles outside a hard-coded list of thirteen bot IDs silently pass through, limiting general installation usefulness.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 152278,
      "name": "phoenix",
      "does": "Journals tool-call intents and outcomes in SQLite and offers explicit crash recovery, optional deduplication and a fault-injection report interface.",
      "useful": 4,
      "reliable": 1,
      "polish": 4,
      "creative": 4,
      "beyond": 5,
      "total": 23,
      "best": "A durable per-tool ledger and native execution middleware target the important ambiguity between a side effect and its recorded result.",
      "worst": "core.py marks an operation recovered whenever any fingerprinted target changed, so a partial or unrelated write can be mistaken for full completion despite the exactly-once claims.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 152813,
      "name": "prevent-this-next-time",
      "does": "Enforces chat-managed block or owner-approval rules on configured connector writes, staging exact requests for one-shot release from a Telegram DM.",
      "useful": 3,
      "reliable": 4,
      "polish": 3,
      "creative": 3,
      "beyond": 5,
      "total": 23,
      "best": "Owner identity checks, canonical destinations and crash-resistant approval claims give real enforcement rather than a model’s promise to ask first.",
      "worst": "Preventing alternate write paths requires a dedicated restrictive profile and a separate launcher tied to Hermes 0.23.1 internals, with manual connector declarations.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 150516,
      "name": "delegation-receipts",
      "does": "Requires subagent receipts and checks their file claims and verification commands, delivering adverse verdicts to the parent and keeping a local ledger.",
      "useful": 4,
      "reliable": 4,
      "polish": 3,
      "creative": 4,
      "beyond": 5,
      "total": 25,
      "best": "Checks use the child’s own tool history, decline unproven working directories and keep host command execution off by default.",
      "worst": "It cannot veto completion, file attribution remains heuristic, and proving test success requires opting into unsandboxed host execution of the child’s cited check.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 153084,
      "name": "nyx-wiki-recall",
      "does": "Injects decision-model-selected Markdown wiki glimpses into Hermes turns and writes daily transcripts, optionally serving as a memory provider.",
      "useful": 3,
      "reliable": 3,
      "polish": 3,
      "creative": 3,
      "beyond": 5,
      "total": 22,
      "best": "Native hooks and an optional memory provider combine hierarchical wiki recall with automatic daily transcript capture and extensive edge-case tests.",
      "worst": "Lookups send unredacted conversation context to OpenRouter and can add roughly twenty seconds per turn, while installation needs a documented workaround to avoid replacing the existing memory provider.",
      "standout": false,
      "blocked": null
    }
  ]
}