{
 "batch": "read1-082",
 "plugins": [
  {
   "pr": 0,
   "name": "done-bell",
   "does": "Rings an OS notification and chime, and shows a busy/done/needs-you chip in Desktop, when a Hermes turn finishes or waits for approval.",
   "useful": 3,
   "reliable": 3,
   "polish": 2,
   "creative": 2,
   "beyond": 3,
   "total": 16,
   "best": "Small, defensive code: every hook and tool swallows errors, state is written atomically, and nine local tests cover the hook cycle and mute gate.",
   "worst": "The README is a developer note with the author's own Windows paths in the install command and a manual copy step for the Desktop pane, with no screenshot.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 0,
   "name": "homeassistant",
   "does": "Official Nous plugin that makes Home Assistant a gateway platform (WebSocket state events in, persistent notifications out) and adds four REST tools to list, read and control entities.",
   "useful": 4,
   "reliable": 5,
   "polish": 4,
   "creative": 2,
   "beyond": 5,
   "total": 25,
   "best": "A full platform adapter with reconnect and heartbeat handling, operator allow/deny lists enforced before any network call, secrets read through Hermes's secret scope, and about 2,400 lines of tests.",
   "worst": "It is the former core integration moved out of Hermes, so it brings nothing new as an idea and has no Desktop UI.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 151897,
   "name": "adplug",
   "does": "Portable package of 16 PPC skills that drive AdPlug's hosted MCP server for Google, Meta, LinkedIn, Microsoft and Reddit Ads, which the user must add to Hermes by hand.",
   "useful": 3,
   "reliable": 3,
   "polish": 4,
   "creative": 2,
   "beyond": 1,
   "total": 14,
   "best": "Deep, well-written skill content with per-platform audit and build references, and a preview-then-approve rule for every spend change.",
   "worst": "The package holds no code at all, not even an mcp.json: it is Markdown skills for a paid SaaS, so a skill folder would do exactly the same.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 153750,
   "name": "strategic-arena",
   "does": "Runs a background multi-agent stakeholder simulation over a repo, paper or post through the session's own model and returns a friction score, veto players, priced objections and dated claims scored later in a Desktop pane.",
   "useful": 3,
   "reliable": 4,
   "polish": 4,
   "creative": 4,
   "beyond": 4,
   "total": 23,
   "best": "About 4,300 lines of engine code with 260 tests, model access only through Hermes's plugin LLM facade, friendly provider-error messages and a 2,800-line Desktop pane.",
   "worst": "A niche decision-review tool whose 27,000-character README and committed internal notes (HANDOFF, SESSION-PROMPT, PROGRESS) promise more rigor (Shapley-priced objections, calibration) than a hundred-call LLM debate can deliver, and the PR is still unmerged.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 151929,
   "name": "promise-keeper",
   "does": "Turns a rule the user states in chat into a human-confirmed, deterministic guard that checks every tool call through a tool_execution middleware and pre_tool_call hook, blocking or asking while quoting the user's words.",
   "useful": 4,
   "reliable": 5,
   "polish": 4,
   "creative": 4,
   "beyond": 4,
   "total": 25,
   "best": "Enforcement is deterministic code outside the model, confirmable only by a person, with 130 tests including three red-team rounds against shell, PowerShell, nested-shell and code bypasses, and documented known limits.",
   "worst": "The 2,400-line command-effects parser is the whole product and remains best effort for terminals, so a user may trust a 'protected folder' more than the parser can guarantee.",
   "standout": true,
   "blocked": null
  },
  {
   "pr": 153939,
   "name": "pfwa",
   "does": "Eight stdlib-only tools that draw provably fair random numbers from the drand beacon and read, quote and build unsigned trades for PFWA prediction markets on Robinhood Chain.",
   "useful": 2,
   "reliable": 3,
   "polish": 3,
   "creative": 3,
   "beyond": 3,
   "total": 17,
   "best": "Holds no keys: trades come back as unsigned transactions for the user's own wallet, and the drand draw commits to a future round so the proof can be rechecked.",
   "worst": "No tests at all, and six of eight tools serve one vendor's prediction-market product that few Hermes users will touch.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 152309,
   "name": "no-manual",
   "does": "Helps assemble furniture with no instructions: finds the official manual, reverses disassembly photos into steps, or solves flat-pack carcass furniture from part photos via structured vision calls, then coaches step by step with photo checks and a Desktop 3D step viewer.",
   "useful": 3,
   "reliable": 4,
   "polish": 5,
   "creative": 5,
   "beyond": 4,
   "total": 25,
   "best": "A finished product: 14 tools, a deterministic solver and renderer with golden tests, a fail-closed safety gate for cribs and load-bearing furniture enforced in both handlers and a pre_tool_call hook, a 56-second demo video and a bundled three.js viewer.",
   "worst": "It needs a vision-capable model and a gateway restart for the PDF renderer, and the solver only covers carcass furniture, so real-world coverage is narrower than the banner suggests.",
   "standout": true,
   "blocked": null
  },
  {
   "pr": 150749,
   "name": "grokbot-staff",
   "does": "Adds a chief-of-staff and specialist personas that answer through ctx.llm, a delegation backend for delegate_task, and an optional fail-closed webhook or inbox-folder bridge that hands jobs to the user's own Grok Bot.",
   "useful": 2,
   "reliable": 4,
   "polish": 3,
   "creative": 2,
   "beyond": 3,
   "total": 17,
   "best": "Careful bridge code: https-only webhook stored as a Hermes secret, redirects refused, path-traversal checks on job ids, and 20 tests including a manifest-matches-registration check.",
   "worst": "The specialists are just system prompts a skill could carry, and the Grok Bot half is a generic webhook POST plus a shared folder because Grok Bot has no API, so little of it is specific to Hermes or Grok Bot.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 149686,
   "name": "botsesh",
   "does": "Registers BotSesh shared rooms as a native Hermes gateway platform with OAuth through Hermes auth, SSE room streaming, peer handoffs, nine room tools and DM controls.",
   "useful": 2,
   "reliable": 3,
   "polish": 3,
   "creative": 3,
   "beyond": 5,
   "total": 21,
   "best": "A real platform adapter with 17,000 lines, 154 tests, a security policy and documented recovery paths, and sign-in handled by Hermes rather than a key file.",
   "worst": "The README says this catalog build needs an unmerged Hermes PR (#151209) and that stock Hermes users should install a different version from main, it is Linux-only, and the dense 27,000-character README makes setup hard to follow.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 152419,
   "name": "goalpost",
   "does": "A pre_tool_call hook that blocks write_file, patch and terminal calls which weaken tests (skip markers, deleted tests, removed assertions, || true, --no-verify) and records every decision in a local ledger.",
   "useful": 4,
   "reliable": 4,
   "polish": 3,
   "creative": 3,
   "beyond": 3,
   "total": 20,
   "best": "Small, deterministic and honest: 600 lines, 32 tests, no network or model calls, and a limits section that says exactly what it cannot catch.",
   "worst": "The detector is text counting on a guessed test-file pattern, blocks legitimate test deletion, and the README admits it was not yet tested on Linux, macOS or Docker.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 151336,
   "name": "superlist",
   "does": "Portable package with an mcp.json pointing at Superlist's hosted MCP server plus one skill, so Hermes can manage the user's tasks and lists after OAuth sign-in.",
   "useful": 3,
   "reliable": 3,
   "polish": 3,
   "creative": 1,
   "beyond": 2,
   "total": 14,
   "best": "Marks the server untrusted so Hermes asks before every write, uses OAuth through Hermes with no keys in the package, and the skill teaches find-before-change and confirm-before-delete.",
   "worst": "The whole plugin is a two-line mcp.json and a skill for a vendor's existing MCP server, and the README is the generic multi-client Superlist MCP page rather than a Hermes one.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 150360,
   "name": "dike",
   "does": "One tool that runs a vendored, offline Chilean-law engine in a sandboxed subprocess to check every statute article citation in a legal brief against a hash-verified local corpus of 25 statutes, marking each verified or unverified with a reason.",
   "useful": 2,
   "reliable": 4,
   "polish": 3,
   "creative": 4,
   "beyond": 3,
   "total": 19,
   "best": "Verification is deterministic and auditable: the corpus ships with SHA-256 sums, the engine runs with a minimal environment and no credentials, and the adapter validates the engine's answer against a fixed contract with 50 tests.",
   "worst": "A Linux-only, Spanish-language tool for Chilean lawyers with five Python dependencies, a slow first-call unpack, and a README section left as a placeholder to be completed by the maintainer.",
   "standout": false,
   "blocked": null
  }
 ]
}