{
 "batch": "read1-138",
 "plugins": [
  {
   "pr": 0,
   "name": "done-bell",
   "does": "Notifies the user (OS toast, chime, Desktop status chip) when a Hermes turn ends or is waiting on an approval.",
   "useful": 3,
   "reliable": 3,
   "polish": 2,
   "creative": 2,
   "beyond": 3,
   "total": 16,
   "best": "Hooks the approval request so the user learns Hermes is blocked on them, with no keys and no stored command text.",
   "worst": "Thin README with Windows-only install from the author's own path, unused config_schema, a /bell command that returns raw JSON and a placeholder sha in docs/catalog.yaml.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 0,
   "name": "homeassistant",
   "does": "Adds Home Assistant as a Hermes gateway platform (events in, notifications out) plus four tools to list, read and call device services.",
   "useful": 5,
   "reliable": 5,
   "polish": 4,
   "creative": 2,
   "beyond": 5,
   "total": 26,
   "best": "About 176 tests, secrets read through Hermes's profile-scoped secret store, and a blocklist of code-execution and SSRF service domains.",
   "worst": "It moved out of Hermes core rather than being new work, and ha_call_service can act on locks or alarms without a per-call approval unless the operator sets deny lists.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 151384,
   "name": "stave",
   "does": "A Hermes Desktop page where the agent writes checked ABC sheet music from a sentence, plays it with note highlighting, revises selected bars and exports PDF/SVG/MIDI/WAV.",
   "useful": 3,
   "reliable": 4,
   "polish": 5,
   "creative": 5,
   "beyond": 4,
   "total": 25,
   "best": "A real code checker (bar lengths, range, syntax) validates the model's notation, inside a finished Desktop page with GIFs, its own model slot and 48 Python/JS tests.",
   "worst": "A niche audience, and stave_open casts tempo with int() without error handling.",
   "standout": true,
   "blocked": null
  },
  {
   "pr": 154039,
   "name": "hive",
   "does": "Learns from exit codes which commands fixed a failing terminal command and appends a 'fixed before' hint to repeat errors, optionally sharing fixes between installs over Nostr.",
   "useful": 4,
   "reliable": 3,
   "polish": 4,
   "creative": 5,
   "beyond": 4,
   "total": 24,
   "best": "Evidence-based fix learning (flaky retries ignored) with egress redaction, signed events, peer quorums and 22 offline tests against a mock relay.",
   "worst": "Peer hints go into tool output the agent reads, defended mainly by cheap proof-of-work and a regex blocklist, and the read-mode relay lookup runs synchronously under a lock on the hot path.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 153182,
   "name": "antigravity-acp-provider",
   "does": "A model provider that runs Google's official antigravity-acp server as a child process and talks ACP JSON-RPC over stdio, with a pinned binary hash and a warm process pool.",
   "useful": 3,
   "reliable": 2,
   "polish": 3,
   "creative": 3,
   "beyond": 5,
   "total": 21,
   "best": "A real new model backend for Hermes with about 729 tests against a fake ACP server and CI on Linux and Windows.",
   "worst": "It auto-approves the Google agent's native file edits and shell commands, bypassing Hermes terminal approval with no switch, and its own status table says real auth and inference are unverified for this release.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 150164,
   "name": "crashlab",
   "does": "Injects lost-response and restart faults into a web app through a real Chromium, freezes the failing run as an experiment and re-runs it after Hermes repairs the app.",
   "useful": 3,
   "reliable": 4,
   "polish": 4,
   "creative": 5,
   "beyond": 4,
   "total": 24,
   "best": "Approvals bound to a digest of the exact plan, about 311 tests, and recorded before/after evidence on outside apps (Kist, Flaskr).",
   "worst": "A dense, repetitive README and 14 tools aimed at a narrow web-reliability audience.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 153617,
   "name": "memory-ferry",
   "does": "A local SQLite memory provider that migrates MEMORY.md and USER.md through a dry-run, digest-approved import, verify and rollback flow.",
   "useful": 2,
   "reliable": 3,
   "polish": 2,
   "creative": 2,
   "beyond": 4,
   "total": 17,
   "best": "Careful migration: the approval fails if the source changed after planning, and rollback deletes only records it can prove it imported.",
   "worst": "Search is plain word overlap, the 'retrieval parity' check runs the same ranker on both sides, and the plugin.yaml is four lines with no screenshots or demo.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 153778,
   "name": "nerdit",
   "does": "Deploys folders the agent builds as supervised local Docker services with URLs through the Nerdit engine and warns the agent when one goes down.",
   "useful": 4,
   "reliable": 3,
   "polish": 3,
   "creative": 4,
   "beyond": 4,
   "total": 22,
   "best": "Gives Hermes persistent self-hosting with outage alerts, a pre_tool_call guard and a CI-run test against a fake engine.",
   "worst": "The guard is a fixed name list that misses stop_service and pass-through nerdit_call, the child engine inherits every host env var including API keys, and an injected hint steers the agent away from other hosts.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 153861,
   "name": "summon",
   "does": "A gateway platform that turns a second X account into the user's own reply bot: tag it under a post and their Hermes reads the thread and replies publicly.",
   "useful": 4,
   "reliable": 4,
   "polish": 5,
   "creative": 5,
   "beyond": 5,
   "total": 28,
   "best": "Owner-only triggers, keys saved and read through Hermes's secret store, a daily budget, dry-run mode and a secret/contact filter on replies, plus 32 tests and a relay e2e.",
   "worst": "It posts publicly with no per-reply approval, and the free and Borrow modes depend on the author's hosted relay, which holds the bot's X tokens.",
   "standout": true,
   "blocked": null
  },
  {
   "pr": 152251,
   "name": "aunx-decompose",
   "does": "A planning skill where the agent writes a task as typed units with data edges, and a Python checker validates it and prints parallel waves and the critical path.",
   "useful": 2,
   "reliable": 2,
   "polish": 3,
   "creative": 3,
   "beyond": 2,
   "total": 14,
   "best": "A real stdlib checker verifies the plan's structure, with 62 tests on the checker.",
   "worst": "The Hermes side is a 40-line wrapper whose /decompose only returns 'load the skill' text, none of the tests touch it, and the README says only the Claude Code plugin is packaged.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 151984,
   "name": "customs",
   "does": "Plants unique honeytoken credential files and blocks or flags any agent tool call that reaches for them, redacting the fake values from tool results and outgoing LLM requests.",
   "useful": 4,
   "reliable": 3,
   "polish": 4,
   "creative": 5,
   "beyond": 4,
   "total": 24,
   "best": "A new defence that catches an agent hunting for secrets inside the tool loop, with attack-chain correlation, session lockdown and a candid limitations section.",
   "worst": "No pytest suite (only a 10-sample script and an in-plugin self-check) for about 4.3k lines, and every hook fails open by design.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 150556,
   "name": "hermes-ui-cards",
   "does": "Lets the agent answer with 17 kinds of interactive card (plan, compare, form, quiz and more) rendered in Hermes Desktop and the Webapp, with the user's answer returned as their next message.",
   "useful": 4,
   "reliable": 4,
   "polish": 5,
   "creative": 4,
   "beyond": 5,
   "total": 27,
   "best": "It changes what Hermes Desktop can show, with about 135 Python and 197 jsdom tests, a real-Hermes e2e and screenshots of every card.",
   "worst": "Tested only against Hermes 0.22.7, answers.db never expires, and the README has an unfilled demo-video placeholder.",
   "standout": true,
   "blocked": null
  }
 ]
}