{
  "batch": "read1-073",
  "plugins": [
    {
      "pr": 0,
      "name": "done-bell",
      "does": "Notifies users when Hermes finishes a turn or needs approval, with a Desktop status chip, settings pane and CLI alerts.",
      "useful": 4,
      "reliable": 3,
      "polish": 3,
      "creative": 2,
      "beyond": 4,
      "total": 20,
      "best": "Hooks and Desktop busy-state observation solve a common unattended-turn annoyance without credentials.",
      "worst": "Installation still includes a developer-specific Windows path and manual Desktop copying, while tests cover backend state rather than actual notification delivery.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 0,
      "name": "homeassistant",
      "does": "Connects Hermes to Home Assistant through device tools, 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 reactive smart-home events, cron delivery, scoped secrets and carefully tested service-target restrictions.",
      "worst": "Entity restrictions are opt-in, so unrestricted installations still permit broad device control; the plugin has no dedicated Desktop control interface.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 153512,
      "name": "equitylayer",
      "does": "Tracks research theses and sector readings using EquityLayer MCP, local baselines, scheduled checks and a local HTML dashboard.",
      "useful": 3,
      "reliable": 4,
      "polish": 4,
      "creative": 3,
      "beyond": 4,
      "total": 22,
      "best": "Persistent research baselines and cron-aware acknowledgment prevent unattended checks from silently dismissing information needing a person.",
      "worst": "Its usefulness depends on one external research service and narrow sector coverage, with some advertised themes still missing live readings.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 153714,
      "name": "tenure",
      "does": "Adds revocable per-action autonomy tiers, approval hooks, undo snapshots, model-specific evidence and a signed local audit passport.",
      "useful": 5,
      "reliable": 4,
      "polish": 5,
      "creative": 5,
      "beyond": 5,
      "total": 29,
      "best": "Deep integration with approvals, model changes and subagent taint combines real undo support with unusually extensive regression tests and finished multilingual interfaces.",
      "worst": "Its static command classifier is not a sandbox and broad trusted action types can miss dangerous effects; it also relies on private Hermes imports.",
      "standout": true,
      "blocked": null
    },
    {
      "pr": 152126,
      "name": "unbl0ck",
      "does": "Collects live prompts, pairing requests, blocked Kanban cards and unanswered chats into a Desktop queue with human-triggered responses.",
      "useful": 5,
      "reliable": 3,
      "polish": 4,
      "creative": 4,
      "beyond": 5,
      "total": 26,
      "best": "A single actionable queue addresses the practical bottleneck of supervising multiple Hermes runs while keeping decisions with the person.",
      "worst": "Queue detection uses language heuristics and private Kanban/database interfaces, while automatic summaries send cross-profile item text to the configured model by default.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 149934,
      "name": "shelly-direct",
      "does": "Discovers and controls Shelly devices directly on the LAN with read-back verification and a web dashboard.",
      "useful": 3,
      "reliable": 4,
      "polish": 4,
      "creative": 3,
      "beyond": 4,
      "total": 22,
      "best": "Generation-aware device control, bounded private-network discovery and stateful device-stub tests make the hardware integration concrete.",
      "worst": "The chat settings loader omits allow_model_sweep, so enabling that documented setting never reaches the handler through ctx.get_config.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 152144,
      "name": "hermes-security-plugin",
      "does": "Connects Hermes to remote security-header checks and penetration-testing provider lookups through two MCP servers and a workflow skill.",
      "useful": 3,
      "reliable": 3,
      "polish": 3,
      "creative": 2,
      "beyond": 3,
      "total": 17,
      "best": "The tiny package clearly discloses its remote data flows and marks both servers untrusted for Hermes approval handling.",
      "worst": "No checking implementation or tests are pinned locally, so reliability depends on mutable remote tools and a skill rather than a reviewed security engine.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 149464,
      "name": "asin-hhc-proof-of-pay",
      "does": "Validates ASIN-HHC receipt structure, checks a published provenance hash chain and reads a work-market board snapshot.",
      "useful": 2,
      "reliable": 3,
      "polish": 3,
      "creative": 3,
      "beyond": 3,
      "total": 17,
      "best": "Read-only local schema and chain checks offer concrete verification while explicitly distinguishing internal consistency from independent proof.",
      "worst": "The shipped test harness looks for a nonexistent nested asin-hhc-proof-of-pay package, and chain consistency cannot independently establish that the work or payment occurred.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 152358,
      "name": "exchangerebel",
      "does": "Adds an ExchangeRebel remote MCP connection and a skill for free discovery and paid x402 API calls.",
      "useful": 3,
      "reliable": 2,
      "polish": 3,
      "creative": 2,
      "beyond": 3,
      "total": 16,
      "best": "A minimal credential-free connector marks its server untrusted and explains paid-call prices and approval expectations.",
      "worst": "The package supplies no payment implementation or end-to-end tests, and its README still describes header/query payment retries while the pinned skill specifies MCP payment metadata.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 149142,
      "name": "odoo-prod-guard",
      "does": "Provides Odoo JSON-2 tools and a production-write approval hook with a routine allowlist and gate-history pane.",
      "useful": 3,
      "reliable": 2,
      "polish": 3,
      "creative": 3,
      "beyond": 4,
      "total": 19,
      "best": "A native pre-tool approval hook makes production policy enforceable beyond a prompt, and credentials use Hermes secret scope.",
      "worst": "It defaults to ungated dev mode even with live credentials, and interpolates unchecked model/method strings into request URLs rather than validating that the executed endpoint matches the gate decision.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 154050,
      "name": "gnothi",
      "does": "Recomputes quantitative claims and statistical comparisons, qualifies conclusions and optionally appends narrow automatic audit warnings.",
      "useful": 3,
      "reliable": 4,
      "polish": 3,
      "creative": 3,
      "beyond": 3,
      "total": 19,
      "best": "Deterministic local checks distinguish missing evidence from false claims and ship substantial statistical and invalid-input regression coverage.",
      "worst": "Conclusion checking is pattern matching over unauthenticated supplied evidence; automatic auditing only recognizes tightly labeled paired-binary comparisons, leaving most ordinary agent claims outside its scope.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 152563,
      "name": "computer-view",
      "does": "Provides an authenticated mobile desktop viewer with Telegram entry and coordinated human takeover, handback and fresh-observation enforcement.",
      "useful": 4,
      "reliable": 4,
      "polish": 3,
      "creative": 4,
      "beyond": 5,
      "total": 25,
      "best": "Real RFB input filtering and cross-process control fencing prevent supported agent actions during human control and require a fresh screenshot after handback.",
      "worst": "Setup requires a dedicated Linux Xvnc desktop, private sockets, externally managed HTTPS and explicit in-process plugins; shell and direct desktop clients remain outside its cooperative fence.",
      "standout": false,
      "blocked": null
    }
  ]
}