{
 "batch": "read1-030",
 "plugins": [
  {
   "pr": 0,
   "name": "done-bell",
   "does": "Rings an OS notification/chime and shows a busy/done/needs-you chip when a Hermes turn finishes, with /bell commands and a Desktop pane.",
   "useful": 3,
   "reliable": 4,
   "polish": 4,
   "creative": 3,
   "beyond": 3,
   "total": 20,
   "best": "Small, clean, no keys, Desktop chip and desktop pane, tests folder, avoids storing approval commands.",
   "worst": "Narrow convenience; a notification is close to what hooks alone give, and the README documents no test command.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 0,
   "name": "homeassistant",
   "does": "Official gateway platform plus four REST tools (list/get/service call) for Home Assistant, with entity allow/deny lists and blocked risky domains.",
   "useful": 5,
   "reliable": 4,
   "polish": 4,
   "creative": 3,
   "beyond": 4,
   "total": 24,
   "best": "Puts Hermes in the home: event-bus gateway platform plus guarded tools, with domain/entity validation and tests.",
   "worst": "Token is read from ~/.hermes/.env per README rather than clearly from the secret store; the token itself is not scoped.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 149491,
   "name": "delegation-scout",
   "does": "Delegation backend that routes each delegate_task run to a model route using task heuristics and recorded outcomes in SQLite, with a scout_stats tool.",
   "useful": 3,
   "reliable": 4,
   "polish": 3,
   "creative": 4,
   "beyond": 4,
   "total": 22,
   "best": "Changes how Hermes delegates (a backend plus outcome learning), stdlib only, fails closed with no routes, has unit and e2e tests.",
   "worst": "Needs hand-written routes config; benefit is hard to see for most users and the heuristics are keyword-based.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 151035,
   "name": "sogni",
   "does": "Adds Sogni's decentralised GPU network as an image_gen provider by driving the vendor CLI, with four models and image editing.",
   "useful": 3,
   "reliable": 3,
   "polish": 3,
   "creative": 2,
   "beyond": 3,
   "total": 17,
   "best": "Reads SOGNI_API_KEY via agent.secret_scope.get_secret, so it follows the catalog's secret-store rule.",
   "worst": "No tests seen, needs a global npm CLI install, and there is no spend cap on billed jobs.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 154002,
   "name": "hermes-studiotwin",
   "does": "Connects Hermes and Desktop to the StudioTwin remote MCP for 3D, material, rig, motion and SFX generation, with a Desktop workspace and helper scripts.",
   "useful": 3,
   "reliable": 3,
   "polish": 4,
   "creative": 3,
   "beyond": 3,
   "total": 19,
   "best": "Finished product feel: Desktop workspace, per-tool cost estimates for the approval UI, key kept in the secret store, tests and docs.",
   "worst": "Paid generation with no session spend cap, a private repo, Chromium run with --no-sandbox, and the tool list is from a fixture snapshot; mostly a vendor MCP wrapper.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 152355,
   "name": "recalllayer",
   "does": "Local-first SQLite memory provider with working, episodic and semantic layers, five tools, auto recall and capture, and a CLI.",
   "useful": 4,
   "reliable": 3,
   "polish": 3,
   "creative": 3,
   "beyond": 4,
   "total": 21,
   "best": "A real native memory provider that changes what Hermes remembers, with layers, hook mapping and a mirror of builtin memory.",
   "worst": "Keyword-only FTS recall, auto-capture with little filtering, daemon threads, and the README lists only a store unit test.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 154012,
   "name": "delivery-shelf",
   "does": "Saves files Hermes delivers, groups them into versioned deliveries, and offers /shelf commands, a Desktop Saved page and a Telegram panel.",
   "useful": 3,
   "reliable": 3,
   "polish": 4,
   "creative": 3,
   "beyond": 3,
   "total": 19,
   "best": "Broad surface (Desktop, Telegram, chat commands) with changelog, security policy and honest scope limits.",
   "worst": "Selection of what to save is manual (/shelf add), file-type and size limits apply, and I could not confirm test coverage or hooks from the docs.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 149553,
   "name": "hermes-time-machine",
   "does": "Adds approval-gated irreversible actions, an effect ledger with compensation commands, effect memory and agent-visible checkpoints (rewind/fork) to Hermes.",
   "useful": 4,
   "reliable": 4,
   "polish": 5,
   "creative": 5,
   "beyond": 5,
   "total": 28,
   "best": "Closes a real gap that /rollback cannot: it gates and records git push-style effects, re-informs the model, and checkpoints files, through five hooks, five tools and a slash command.",
   "worst": "Fail-open hooks, filesystem-only scope, Windows-only benchmark numbers, and no compensation command for many effect types.",
   "standout": true,
   "blocked": null
  },
  {
   "pr": 152063,
   "name": "undo-send",
   "does": "Holds outward actions (messages, MCP writes, publishing shell commands) for a short grace window so the user can cancel them from Desktop, chat or commands.",
   "useful": 4,
   "reliable": 4,
   "polish": 4,
   "creative": 4,
   "beyond": 5,
   "total": 26,
   "best": "Uses tool_execution middleware to add a cancel window to outbound actions, with Desktop chip, dashboard and chat undo: a human-in-the-loop only a plugin can supply.",
   "worst": "Fail-open if the plugin errors, adds latency to cron runs, and only 3 commits with unverified privacy claims.",
   "standout": true,
   "blocked": null
  },
  {
   "pr": 150185,
   "name": "tenderlead-agent-plugin",
   "does": "A portable agent-plugin package holding one skill that reviews UK tender documents against a supplier profile to suggest BID or SKIP.",
   "useful": 2,
   "reliable": 3,
   "polish": 3,
   "creative": 2,
   "beyond": 1,
   "total": 12,
   "best": "Honest about scope: no install scripts, hooks, dependencies or credentials, and a clear vendor-published manifest.",
   "worst": "It is a single skill with no code, tools or tests, and is aimed at a vendor product, so it adds nothing a prompt could not do.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 151123,
   "name": "hermes-sysadmin",
   "does": "Manages on-prem Windows and Active Directory through 71 typed tools, a 231-entry vetted PowerShell catalog, tiered approvals, an audit log and a Desktop dashboard.",
   "useful": 3,
   "reliable": 4,
   "polish": 5,
   "creative": 3,
   "beyond": 4,
   "total": 23,
   "best": "Serious engineering: tiers, protected objects, parameter binding, audit log, offline and live test suites, Desktop dashboard.",
   "worst": "Audience is niche (AD admins), and the password for WinRM needs an env passthrough for cron.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 153822,
   "name": "lights-out",
   "does": "Prices AWS CLI creates for approval, ledgers every created resource with its delete command, posts a cost receipt, and /lights out deletes the unkept ones.",
   "useful": 4,
   "reliable": 4,
   "polish": 4,
   "creative": 4,
   "beyond": 5,
   "total": 26,
   "best": "Turns cloud-bill surprise into an approval step plus a cleanup ledger, with 140 offline tests and a benchmark against real Hermes and moto.",
   "worst": "Sees only AWS CLI commands (not boto3, Terraform), prices are us-east-1 list only, and /lights out deletes from a ledger that could be wrong.",
   "standout": true,
   "blocked": null
  }
 ]
}