{
 "batch": "read1-084",
 "plugins": [
  {
   "pr": 0,
   "name": "done-bell",
   "does": "Rings an OS notification and chime and shows a busy/done/needs-you status chip when a Hermes turn finishes or waits for approval.",
   "useful": 4,
   "reliable": 4,
   "polish": 3,
   "creative": 2,
   "beyond": 4,
   "total": 21,
   "best": "Small, fail-safe hooks (session end, approval request/response) plus a Desktop chip and /bell command, with unit tests and no secrets stored.",
   "worst": "README install steps are the author's own Windows dev paths and need a manual copy of the Desktop pane; the idea itself is a simple notifier.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 0,
   "name": "homeassistant",
   "does": "Official Home Assistant gateway platform plus four tools to list, read and call smart-home entities and services over REST/WebSocket.",
   "useful": 5,
   "reliable": 5,
   "polish": 4,
   "creative": 3,
   "beyond": 5,
   "total": 27,
   "best": "A full platform adapter with operator allow/deny lists for service calls, token through Hermes's secret store, and a large test suite.",
   "worst": "Not new as an idea (it was previously built into Hermes core), and the README has no screenshots or demo beyond a header image.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 150667,
   "name": "dropstation",
   "does": "Gateway platform that turns private local folders into Hermes inboxes: each dropped file runs a fresh session under a lane's standing task and writes a Markdown report and SHA-256 receipt to an outbox.",
   "useful": 3,
   "reliable": 4,
   "polish": 5,
   "creative": 4,
   "beyond": 5,
   "total": 26,
   "best": "Carefully built spool: O_NOFOLLOW reads, symlink/FIFO refusal, a worker lease, dedup by content hash and crash recovery, all covered by eight tests.",
   "worst": "Polling-based folder watcher with a one-job-at-a-time worker and only spool-level tests, nothing that exercises the adapter itself; the audience is people who already script file drops.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 149797,
   "name": "bench-rewind",
   "does": "Undo history for a physical device teardown: records removal steps on your own photos with part regions, bags and dependencies, answers 'where does this go?', and plans reassembly in valid reverse order via four tools and a Desktop page.",
   "useful": 3,
   "reliable": 5,
   "polish": 5,
   "creative": 5,
   "beyond": 3,
   "total": 24,
   "best": "Exceptionally finished: ~150 tests (model, store, images, browser UI), metadata stripping on photos, no-follow private store, a screenshot, a demo kit and a script-free printable reassembly packet.",
   "worst": "The agent only drives a record-keeping engine the user fills in by hand; the core value is the standalone workbench rather than a change to what Hermes itself can do, and it ships a 700 KB built web bundle.",
   "standout": true,
   "blocked": null
  },
  {
   "pr": 151063,
   "name": "listing-automater",
   "does": "Skill-only plugin that registers one SKILL.md describing a commercial Facebook Marketplace posting app for real-estate agents and giving listing-writing tips.",
   "useful": 1,
   "reliable": 3,
   "polish": 2,
   "creative": 1,
   "beyond": 1,
   "total": 9,
   "best": "Honest and harmless: the PR and manifest say it is skill only, with no tools, hooks, network or env.",
   "worst": "It is a product advertisement for a paid third-party app delivered as a skill; a prompt or skill does exactly the same and there is no code doing work.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 154065,
   "name": "axiom",
   "does": "Registers four SymPy/SciPy tools to check equation/inequality systems, SI dimensional homogeneity, conservation in ledgers and datasets, plus hooks and middleware that claim to self-repair numeric hallucinations.",
   "useful": 2,
   "reliable": 1,
   "polish": 2,
   "creative": 3,
   "beyond": 2,
   "total": 12,
   "best": "The constraint parser walks the AST into SymPy instead of calling eval, and the dimensional and conservation checks are real code.",
   "worst": "Claims more than the code does: the PR advertises Z3/dReal SMT certificates but the code only runs SciPy SLSQP, the 'self-repair' middleware only acts on an axiom_claim field a model never emits, and the only test is assert True.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 152616,
   "name": "dial-a-hermes",
   "does": "Gateway platform that lets you use Hermes from any feature phone over USSD (Africa's Talking-compatible), with paged 182-char screens, keypad 1/2 exec approvals showing the full command, optional SMS replies, Pidgin menus and a local phone simulator.",
   "useful": 3,
   "reliable": 4,
   "polish": 4,
   "creative": 5,
   "beyond": 5,
   "total": 26,
   "best": "Puts the whole agent, including Hermes's own approval cards (once/deny only), behind a USSD short code with a phone-number allowlist, constant-time callback-token check and keys read through Hermes's scoped secret store.",
   "worst": "Real use needs an Africa's Talking account, a public tunnel and a short code, and the 31 core tests plus e2e script sit outside the pinned plugin folder; the audience outside feature-phone regions is small.",
   "standout": true,
   "blocked": null
  },
  {
   "pr": 150766,
   "name": "subscription-sheriff",
   "does": "Tracks subscriptions and free trials from plain-English messages in plugin state, warns before charges via two scheduled-task blueprints and a /sheriff command, and summarizes spend per currency.",
   "useful": 4,
   "reliable": 4,
   "polish": 4,
   "creative": 2,
   "beyond": 3,
   "total": 20,
   "best": "Fully offline, standard-library code with 31 tests, month-end date rolling, per-currency totals without invented exchange rates, and quiet de-duplicated reminders.",
   "worst": "A familiar subscription-tracker idea; much of it (a list plus a cron reminder) could be approximated with Hermes memory and a scheduled prompt.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 152773,
   "name": "project_content",
   "does": "Experimental local project-content store: 13 tools to initialize a chosen folder, organize sections and units, append version-checked edit events, register media references and edit blocks, plus a separate Desktop organizer.",
   "useful": 2,
   "reliable": 2,
   "polish": 2,
   "creative": 2,
   "beyond": 2,
   "total": 12,
   "best": "Careful file handling: exclusive metadata creation with O_EXCL/O_NOFOLLOW, dir-fd operations and optimistic version conflicts instead of overwrites.",
   "worst": "By its own README the Desktop workspace and the agent tools use different stores and do not interoperate, media playback is not implemented, the code is packed into dense one-line statements that are hard to review, and there are only two tests.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 152065,
   "name": "evaluation-ledger",
   "does": "Stores immutable artifact text in a local SQLite ledger with SHA-256/JSON checks, and (opt-in) records allowlisted usage metadata from API-request and tool-call hooks for per-session usage reports.",
   "useful": 2,
   "reliable": 3,
   "polish": 2,
   "creative": 2,
   "beyond": 3,
   "total": 15,
   "best": "Privacy-minded design: usage collection is off unless observe_usage is exactly true, observers keep only an allowlist of ids and counters, and artifacts are insert-only.",
   "worst": "Narrow audience and rough finish: the README still says no public repository is claimed and mentions a 'routing plugin' from its sibling listings, the code is written in dense one-line style, and its tests live outside the pinned plugin folder.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 149269,
   "name": "constraint-vault",
   "does": "Pins exact user constraints (numbers, addresses, hard rules) in a local JSON store and re-injects them verbatim every turn via pre_llm_call and again right after compaction via transform_tool_result, logging a SHA-256 of each injection.",
   "useful": 3,
   "reliable": 3,
   "polish": 3,
   "creative": 3,
   "beyond": 4,
   "total": 20,
   "best": "A targeted fix for a real failure (compaction paraphrasing exact values) using hooks only a plugin has, with a hash-logged delivery record and size caps.",
   "worst": "Only three tests, failures are swallowed silently (a failed save is dropped with no error), the injection log grows without bound, and an agent can pin text from untrusted tool output that then rides every turn.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 151642,
   "name": "ciphervault",
   "does": "Wraps the CipherVault Rust CLI to give Hermes seven secret-vault tools (masked scan, approval-gated RAM-only run, encrypted push, recovery drill, redact, audit), leak-guard and result-scrubbing hooks, a Hermes secret source and a Desktop pane.",
   "useful": 3,
   "reliable": 4,
   "polish": 4,
   "creative": 4,
   "beyond": 5,
   "total": 25,
   "best": "Deep, safety-first integration: every ciphervault_run goes through Hermes's approval gate, other tools' results are scrubbed before entering context, subprocesses get an allow-listed env, and the binary bootstrap is pinned to a SHA-256.",
   "worst": "Depends on a separately built Rust binary (downloaded on a user-run setup step) and a large external project, so setup is heavy and the plugin's guarantees rest on code outside its folder.",
   "standout": false,
   "blocked": null
  }
 ]
}