{
 "batch": "read1-051",
 "plugins": [
  {
   "pr": 0,
   "name": "done-bell",
   "does": "Rings an OS notification/chime and shows a busy/done/needs-you chip when a Hermes turn ends or an approval is waiting, with a /bell command, tool, dashboard routes and a Desktop pane.",
   "useful": 3,
   "reliable": 3,
   "polish": 2,
   "creative": 2,
   "beyond": 3,
   "total": 16,
   "best": "Small, honest, keyless code that hooks on_session_end and approval hooks and reaches CLI, dashboard and Desktop surfaces.",
   "worst": "Rough setup: the README installs from the author's own Windows path, the Desktop pane has to be copied by hand, config_schema is never read and docs/catalog.yaml still holds a placeholder sha.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 0,
   "name": "homeassistant",
   "does": "Nous's official Home Assistant gateway platform plus four REST tools (list entities, get state, list services, call service) for controlling a smart home from Hermes.",
   "useful": 4,
   "reliable": 5,
   "polish": 4,
   "creative": 2,
   "beyond": 5,
   "total": 25,
   "best": "Mature code with about 176 tests, scoped secret lookup for HASS_TOKEN, timeouts everywhere and a strict service blocklist and regex validation.",
   "worst": "A port of formerly built-in core code rather than anything new, and ha_call_service can unlock or disarm devices with no per-call approval beyond the blocklist.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 152971,
   "name": "detonator",
   "does": "Runs another Hermes plugin at a pinned commit inside a locked-down, no-network Docker sandbox with decoy secrets and a clock jump, then reports undeclared behaviour against deterministic rules mapped to MITRE ATT&CK/ATLAS.",
   "useful": 3,
   "reliable": 4,
   "polish": 5,
   "creative": 5,
   "beyond": 3,
   "total": 23,
   "best": "A new kind of agent tool: dynamic detonation with audit hooks, stub imports, time-bomb clock shift and rug-pull diffing, gated by ctx.ask_choice approval and about 83 tests.",
   "worst": "Needs Docker on the user's machine, and its headline catalog-wide run and malicious-fixture results rest on private data that cannot be checked from the repo.",
   "standout": true,
   "blocked": null
  },
  {
   "pr": 152224,
   "name": "twitch-assist",
   "does": "A Twitch chat platform adapter that lets allowlisted viewers !ask the streamer's Hermes agent and lets mods run a Backseat chat vote whose winner becomes the agent's next task.",
   "useful": 3,
   "reliable": 4,
   "polish": 4,
   "creative": 4,
   "beyond": 4,
   "total": 23,
   "best": "Real stdlib IRC-over-TLS platform with a careful trust model (server tags decide who votes, pinned toolsets, cooldowns, notice filtering) and 87 tests against a fake IRC server.",
   "worst": "Public-channel safety still depends on the user setting up a dedicated profile correctly, and the pre_tool_call guard allows calls if session context cannot be imported.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 154070,
   "name": "vivarium",
   "does": "An artificial-life simulation in which every agent tool call feeds a 48x48 dish and every learned skill hashes into a new species, with renders, a codex, gallery and /vivarium command.",
   "useful": 2,
   "reliable": 2,
   "polish": 5,
   "creative": 5,
   "beyond": 3,
   "total": 20,
   "best": "Highly original and very polished, with images, clips, PDFs and about 370 tests.",
   "worst": "Intrusive for a toy: pre_llm_call injects text telling the agent to relay its card verbatim, a polling thread starts at import in every process, and it reads parent processes' /proc environ.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 153010,
   "name": "nousviz",
   "does": "Connects Hermes to self-hosted NousViz analytics servers with 15 tools for browsing the catalog, running SQL and building dashboards, alerts and annotations.",
   "useful": 3,
   "reliable": 2,
   "polish": 3,
   "creative": 3,
   "beyond": 3,
   "total": 17,
   "best": "Practical multi-server REST client with 20s timeouts that keeps the NousViz password in Hermes's secret store.",
   "worst": "No tests, and feedback.py hard-codes a third-party collector URL and token that receive user notes and context by default, contrary to the README's claim that it calls only the configured base_url.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 151253,
   "name": "peakflo-spaces",
   "does": "A config-only package that registers Peakflo's hosted Spaces MCP server (via npx mcp-remote, marked untrusted) and a skill for finance and ops workflows.",
   "useful": 2,
   "reliable": 3,
   "polish": 3,
   "creative": 1,
   "beyond": 2,
   "total": 13,
   "best": "Honest and minimal: it says it ships no code, pins mcp-remote and marks the server untrusted so Hermes asks before write tools.",
   "worst": "No engineering of its own; all value sits in a closed commercial hosted service and Hermes can already add an MCP server without a plugin.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 153878,
   "name": "wiretap",
   "does": "Records the exact request and response of every model API call into a local SQLite store and shows them in a dashboard tab, a Desktop page and a read-only CLI, including cache-miss analysis.",
   "useful": 3,
   "reliable": 4,
   "polish": 5,
   "creative": 4,
   "beyond": 4,
   "total": 24,
   "best": "Deep, honest engineering for debugging Hermes: wire capture beside the transcript, credential redaction, 0600 storage and about 229 Python tests plus browser tests.",
   "worst": "Relies on fragile internals (frame inspection, a process-wide httpx patch, private Hermes imports), stores full conversation content by default, and its manifests still mention unshipped features.",
   "standout": true,
   "blocked": null
  },
  {
   "pr": 151708,
   "name": "hermes-arcade",
   "does": "Three neon canvas games (Snake, Breakout, Pong) in a dashboard tab with a shared high-score store the agent can query through three read-only tools.",
   "useful": 2,
   "reliable": 2,
   "polish": 3,
   "creative": 3,
   "beyond": 2,
   "total": 14,
   "best": "A fun, working, hand-written arcade that uses dashboard, tools and slash command around one atomic score store.",
   "worst": "No tests, the agent side is three trivial leaderboard queries, POST /scores accepts any score, and the README embeds none of its screenshots or demo gif.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 152481,
   "name": "argus",
   "does": "Groups agent turns into intents, has a model review each skill edit Hermes made against evidence, proposes new rules, and applies or rolls back skills only after a person decides in its dashboard.",
   "useful": 3,
   "reliable": 4,
   "polish": 4,
   "creative": 5,
   "beyond": 5,
   "total": 26,
   "best": "A novel human-gated governance layer over Hermes's self-learning skills loop, with strict typing, CI, about 75 tests and apply refusing without a matching recorded decision.",
   "worst": "Expensive and complex: derive runs automatically after idle and spends model tokens, it leans on undocumented Hermes internals, and the docs are dense.",
   "standout": true,
   "blocked": null
  },
  {
   "pr": 150707,
   "name": "oddsradar",
   "does": "Read-only Polymarket odds search and a watchlist with move and threshold alerts checked by a no-agent cron script that costs zero tokens.",
   "useful": 3,
   "reliable": 4,
   "polish": 4,
   "creative": 3,
   "beyond": 3,
   "total": 20,
   "best": "Clever zero-token cron design with a deterministic, well-tested alert engine (25 offline tests, CI), host allowlist, timeouts and no keys or trading.",
   "worst": "Modest scope, and the watchlist claimed to be shared safely across chat, CLI and cron is guarded only by a threading lock, so separate processes can lose updates.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 151482,
   "name": "openinvest-hosted",
   "does": "Could not be read: the pinned repository is unavailable.",
   "useful": 0,
   "reliable": 0,
   "polish": 0,
   "creative": 0,
   "beyond": 0,
   "total": 0,
   "best": null,
   "worst": null,
   "standout": false,
   "blocked": "The repository github.com/longsizhuo/openinvest-site returns 404 and git fetch of the pinned commit failed, so the plugin could not be read."
  }
 ]
}