{
 "batch": "read1-086",
 "plugins": [
  {
   "pr": 0,
   "name": "done-bell",
   "does": "Rings an OS notification and optional chime when a Hermes turn finishes or an approval is waiting, with a busy/done status chip in Desktop and a /bell command.",
   "useful": 3,
   "reliable": 4,
   "polish": 3,
   "creative": 3,
   "beyond": 3,
   "total": 19,
   "best": "Small, readable hooks (on_session_end, pre_approval_request) wrapped so they never raise, plus a Desktop pane, a slash command and tests.",
   "worst": "README install steps hard-code the author's Windows path and require hand-copying the Desktop pane, and the scope is a single convenience notification.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 0,
   "name": "homeassistant",
   "does": "Official Home Assistant integration: a gateway platform that forwards HA WebSocket state changes to the agent plus four REST tools to list, read and control devices.",
   "useful": 5,
   "reliable": 5,
   "polish": 4,
   "creative": 3,
   "beyond": 5,
   "total": 27,
   "best": "Large test suite (about 2,400 test lines) plus entity allow/deny lists, a token declared as a password-type requires_env, and a full gateway platform adapter.",
   "worst": "It is a port of a long-standing built-in integration, so it is not a new idea, and the README has no screenshots or demo.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 151266,
   "name": "ambassly",
   "does": "Packages Ambassly's hosted affiliate-program MCP server (Bearer key) with one workflow skill for creating programs, links and reviewing commissions.",
   "useful": 2,
   "reliable": 3,
   "polish": 3,
   "creative": 2,
   "beyond": 2,
   "total": 14,
   "best": "Honest disclosure: it marks the server untrusted so Hermes asks before write calls, and its skill tells the agent to get a person's decision on each commission.",
   "worst": "There is no plugin code, only an MCP config and a skill for one commercial SaaS with UTM-tagged links. The tool list comes from the vendor's server and the pinned commit does not cover it.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 150254,
   "name": "aegis-gate",
   "does": "Registers a fail-closed pre_tool_call hook that scans every tool call's arguments for EIP-712 payloads, contract execution (checked on Sourcify) and injected instructions, then blocks it, escalates it to human approval or allows it.",
   "useful": 3,
   "reliable": 4,
   "polish": 4,
   "creative": 4,
   "beyond": 4,
   "total": 23,
   "best": "A real guard that only a plugin can add: it classifies calls by content, caps its time budget inside Hermes's callback timeout, escalates rather than allows on error, and ships per-gate tests and an SSRF fetch guard.",
   "worst": "Its value is mostly for agents that touch EVM wallets, its injection scoring is heuristic, and as its README admits, a WARN is auto-approved on unattended surfaces (cron, --yolo).",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 152987,
   "name": "benchline",
   "does": "One read-only tool that preflights a numeric lab-result CSV against a user-supplied analyte-to-unit map and returns a row-referenced QC summary without identifiers or values.",
   "useful": 2,
   "reliable": 4,
   "polish": 3,
   "creative": 2,
   "beyond": 2,
   "total": 15,
   "best": "Carefully bounded: 1 MiB/5,000-row cap, rejects symlinks, no network or writes, never echoes identifiers or measured values, and a 177-line validator test suite.",
   "worst": "A very narrow lab-operator niche, and the work is a CSV check that a skill with a short script could largely do.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 152879,
   "name": "understory",
   "does": "Catalog entry titled 'the family tree of music labels'; the plugin's code could not be read.",
   "useful": 0,
   "reliable": 0,
   "polish": 0,
   "creative": 0,
   "beyond": 0,
   "total": 0,
   "best": "Not assessable.",
   "worst": "Not assessable.",
   "standout": false,
   "blocked": "The pinned repository github.com/projectmehari/understory returns 404 (gone or private), so the commit could not be fetched."
  },
  {
   "pr": 152159,
   "name": "genui",
   "does": "Gives the agent live UI cards in Hermes Desktop (pull request, task, review list, study) that it shows early, updates in place with JSON Merge Patch after each tool result, and uses for button decisions that return to the agent as a turn.",
   "useful": 4,
   "reliable": 4,
   "polish": 5,
   "creative": 5,
   "beyond": 5,
   "total": 28,
   "best": "It changes how Hermes Desktop talks to the user. Schema-validated props come back to the model as fixable errors, cards render in sandboxed frames (allow-scripts only), themes are pluggable, and it ships screenshots and a benchmark against plain Hermes.",
   "worst": "The plugin folder has no tests of its own (tests sit at the repo root beside the benchmark and demo tooling), it adds a turn hint of a few hundred tokens to every Desktop turn, and it relies on the agent choosing to use the cards.",
   "standout": true,
   "blocked": null
  },
  {
   "pr": 152156,
   "name": "claw-messenger",
   "does": "Gateway platform adapter that lets Hermes send and receive iMessage, RCS and SMS (including groups and attachments) through the paid Claw Messenger WebSocket service, plus a create-group tool.",
   "useful": 4,
   "reliable": 4,
   "polish": 4,
   "creative": 3,
   "beyond": 5,
   "total": 25,
   "best": "A full platform adapter with a sender allowlist, the API key declared as a password requires_env and read through Hermes's env helper, https/wss-only enforcement and eight test files.",
   "worst": "Every message goes through a third-party paid relay on onrender.com, and the pip install path tells users to append the key to ~/.hermes/.env by hand.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 149929,
   "name": "bankbridge",
   "does": "Packages the BankBridge hosted MCP server URL (OAuth sign-in) and one skill so Hermes can answer questions about the user's own US bank, card and brokerage accounts at $5/month per bank.",
   "useful": 3,
   "reliable": 3,
   "polish": 4,
   "creative": 2,
   "beyond": 2,
   "total": 16,
   "best": "Read-only by design: no executable code or stored credentials, OAuth through Hermes's own mcp login, and a skill that refuses money-moving requests.",
   "worst": "It is a multi-agent marketing package for a paid SaaS (Claude, Cursor and Gemini manifests): the Hermes side is only a URL plus a skill, and the read-only promise is enforced by the vendor's server, not by any code at this commit.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 150859,
   "name": "us-air-rules",
   "does": "Seven tools that screen a plant or household against EPA Title 40 rules using live eCFR and Federal Register data and bundled rule packs, returning applies/check/skip verdicts with citations, plus scenario compare and state SIP lookup.",
   "useful": 2,
   "reliable": 4,
   "polish": 4,
   "creative": 4,
   "beyond": 3,
   "total": 20,
   "best": "Deep, citation-backed domain engine that uses only the standard library, needs no API key, caps its cache and Federal Register calls, has about 1,400 lines of offline tests and CI, and states clearly that it is not a compliance determination.",
   "worst": "The audience is a small niche (cement and lime plants, EPA air rules), and the plant layer covers only a few industries.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 151704,
   "name": "guild",
   "does": "Shares one human-reviewed skill library across a team's Hermes agents through a git repo: skills are scrubbed for secrets and proposed on branches, approved only from the terminal, TUI or Desktop, and synced back read-only, with a pre_tool_call hook that blocks local edits to team skills.",
   "useful": 4,
   "reliable": 4,
   "polish": 4,
   "creative": 5,
   "beyond": 5,
   "total": 27,
   "best": "It extends Hermes's own self-written skills to whole teams. Agents can propose but never approve, proposals go through Hermes's approval prompt and secret redactor, and hooks keep team skills read-only.",
   "worst": "It needs git and a shared remote, keeps its data in ~/.hermes-guild instead of HERMES_HOME, and has one 237-line test file for a fairly large git workflow.",
   "standout": true,
   "blocked": null
  },
  {
   "pr": 152678,
   "name": "track-record",
   "does": "Logs every reply in which the agent claims a fix worked, checks it against the next test/build exit code and the user's next message, reports hit rate, calibration gap and Brier score per model, and tells the agent its record when its certain claims fall below 80%.",
   "useful": 4,
   "reliable": 4,
   "polish": 3,
   "creative": 5,
   "beyond": 4,
   "total": 24,
   "best": "A new idea that only hooks can do: it measures the user's own agent's honesty and feeds the record back to make it check before claiming, using local pattern matching with no model calls.",
   "worst": "Claim and outcome detection is regex pattern matching, so a vague next user message or an untested claim can be scored wrongly or left unchecked, and the README has no screenshot or demo.",
   "standout": false,
   "blocked": null
  }
 ]
}