{
  "batch": "final-07",
  "ranking": [
    {
      "rank": 1,
      "pr": 149640,
      "name": "hermes-ark",
      "total": 29,
      "why": "U5/R5/P5/C4/B5: Automatic pre-write snapshots, integrity-checked granular restore and owner-confirmed recovery protect the accumulated agent itself, giving Ark the strongest broadly useful reason to stay installed ([pinned source](https://raw.githubusercontent.com/medomohamed22/hermes-ark/a23e529c1f636aa49d974576cdc5cb15fc8242d3/README.md))."
    },
    {
      "rank": 2,
      "pr": 150353,
      "name": "fluent",
      "total": 29,
      "why": "U5/R5/P5/C4/B5: Checking terminal options against the installed program and escalating destructive commands through Hermes approvals improves everyday work with little friction, just behind Ark because it protects a narrower part of the agent ([pinned source](https://raw.githubusercontent.com/Bruno-coder21/fluent/182c4301f2437e784e9022f1f38576e1d68c3f5f/README.md))."
    },
    {
      "rank": 3,
      "pr": 149738,
      "name": "trust-ladder",
      "total": 28,
      "why": "U5/R4/P5/C4/B5: Compiling corrections into persistent tool guards and verification gates addresses a universal agent frustration, ahead of the remaining entries on likely retention, although the matchers remain best-effort rather than a sandbox ([pinned source](https://raw.githubusercontent.com/merdoks/trust-ladder/fae4e9df58767e5e3832c8f65936b97868bc98fc/README.md))."
    },
    {
      "rank": 4,
      "pr": 153844,
      "name": "fine-print",
      "total": 28,
      "why": "U5/R4/P5/C4/B5: Runtime notices and targeted model annotations expose otherwise silent input loss with broad everyday value, but version-specific detectors and fixture-only coverage prevent full reliability marks ([pinned source](https://raw.githubusercontent.com/Joko-F/hermes-fine-print/7b574f1b5c8dc8eafe3ea9f87ae6b0111853f0ba/README.md))."
    },
    {
      "rank": 5,
      "pr": 150644,
      "name": "hermes-canvas",
      "total": 28,
      "why": "U4/R5/P5/C4/B5: File writes and check exit codes drive an interactive plan that users can steer mid-turn, and the real Hermes smoke check refused a fabricated done status, making this a strong everyday coding companion ([pinned source](https://raw.githubusercontent.com/0xbyt4/hermes-canvas/d6eeff0d6f7ec653262537bdd7bb001cc6d3c7f8/README.md))."
    },
    {
      "rank": 6,
      "pr": 150840,
      "name": "kata",
      "total": 28,
      "why": "U4/R4/P5/C5/B5: Mining repeated successful routines into approved deterministic tools is unusually creative capability growth, with rendered-effect rechecks and clear failure traces, though useful routines require history and timed-out dispatches cannot be cancelled ([pinned source](https://raw.githubusercontent.com/opeblow/kata/cc59b08f1bbeeb02524c060faac04edb34de0564/kata/executor.py))."
    },
    {
      "rank": 7,
      "pr": 151887,
      "name": "cmi",
      "total": 27,
      "why": "U5/R4/P4/C4/B5: Reversible context reduction with exact recall and cache-aware selection offers broad value, but the native-engine setup is less straightforward than the leading entries and project memory remains experimental ([pinned source](https://raw.githubusercontent.com/jeremiebergel/context-manager-intelligence/399468f0687c051476961e4bb045eb84263be6b1/README.md))."
    },
    {
      "rank": 8,
      "pr": 150556,
      "name": "hermes-ui-cards",
      "total": 27,
      "why": "U4/R4/P5/C4/B5: Validated interactive answer cards meaningfully expand Desktop conversations and retain text fallbacks, but eventual cross-client locking permits duplicate submissions and keeps reliability below the leaders ([pinned source](https://raw.githubusercontent.com/robbyczgw-cla/hermes-ui-cards/c40df4d41cd5439cf4902995fd4a9e5505483f25/README.md))."
    },
    {
      "rank": 9,
      "pr": 152458,
      "name": "housemate",
      "total": 27,
      "why": "U4/R4/P5/C4/B5: Turning old browser devices into room-aware Hermes speakers with timers and human approvals is an excellent fit for personal hardware, tempered by TLS and speech setup and an audio path I could not exercise live here ([pinned source](https://raw.githubusercontent.com/RaYYeR220/housemate/09c1a1d0a88bfa8762cee6b3ccf94271b0f172c5/README.md))."
    },
    {
      "rank": 10,
      "pr": 150251,
      "name": "gemini-live",
      "total": 27,
      "why": "U4/R4/P5/C4/B5: Full-duplex voice delegates and steers actual Hermes work while keeping the API key server-side and approvals with the person, but provider-specific access and a live audio path not exercised here narrow its appeal ([pinned source](https://raw.githubusercontent.com/cineglobe/hermes-gemini-live/b6c2f24354aeed1a8b5651b62944d14ed96a472d/README.md))."
    },
    {
      "rank": 11,
      "pr": 152908,
      "name": "contract-necromancer",
      "total": 27,
      "why": "U4/R4/P5/C4/B5: Owner-frozen tests, runtime verification and recheckable receipts give coding users a substantive completion gate, below the broader interfaces because its assurance is limited to the frozen contracts and Hermes bounds verification nudges ([pinned source](https://raw.githubusercontent.com/vijaygopalbalasa/hermes-necromancer/0ff7bec40aba10091a84f94025d4471984fd7386/README.md))."
    },
    {
      "rank": 12,
      "pr": 153955,
      "name": "shadow",
      "total": 27,
      "why": "U4/R4/P5/C4/B5: Blind replay comparisons in both answer orders and reversible model switching help users improve their own agent, but paid replay, isolation verification and minimum sample requirements add substantial friction relative to the plugins above ([pinned source](https://raw.githubusercontent.com/akinduroifedayo/shadow/38173bc9c696df3b1f87b5b9b60ccecd36a260ab/README.md))."
    },
    {
      "rank": 13,
      "pr": 150800,
      "name": "hypnos",
      "total": 27,
      "why": "U3/R4/P5/C5/B5: The actual LoRA trainer, promotion exam and fallback router are exceptional agent-level creativity, but the separate training environment, hardware burden and synthetic demonstrations make sustained usefulness less certain for most users ([pinned source](https://raw.githubusercontent.com/0xNoramiya/hypnos/0a809b9085fdeaed63fd909b927afcc41de512de/README.md))."
    },
    {
      "rank": 14,
      "pr": 150596,
      "name": "ariadne",
      "total": 27,
      "why": "U3/R5/P5/C4/B5: Recorded model and tool responses replay the real Hermes loop for upgrade comparison and bisection without rerunning tools, an excellent engineered capability whose continuous audience is mainly developers and regression hunters ([pinned source](https://raw.githubusercontent.com/miracle352/hermes-ariadne/128c95967a0761ad24d3d224441d1502c43bbb69/README.md))."
    },
    {
      "rank": 15,
      "pr": 152691,
      "name": "grimoire",
      "total": 26,
      "why": "U4/R4/P5/C3/B5: A constrained read-only DuckDB engine plus a Desktop data workflow provides substantial practical value for data users, but analytics is a more conventional and narrower improvement than the leading changes to the agent ([pinned source](https://raw.githubusercontent.com/bettersandeep/grimoire/4824773040b131f3179e5896d71d1435908758a7/engine.py))."
    },
    {
      "rank": 16,
      "pr": 153851,
      "name": "foreman",
      "total": 26,
      "why": "U4/R4/P5/C3/B5: Diff-based test-integrity gates and person-only exceptions are valuable coding protection, but best-effort command coverage and the familiar review workflow put it behind the stronger general agent improvements ([pinned source](https://raw.githubusercontent.com/PedValme/foreman/c15453cb2fc07fa674d6f7b548a553b3ce9e6f5d/plugin/README.md))."
    },
    {
      "rank": 17,
      "pr": 153809,
      "name": "living-spaces",
      "total": 26,
      "why": "U4/R3/P5/C4/B5: Persistent sandboxed apps, records and revisions make Desktop substantially more capable, but the pinned package explicitly uses private cron APIs and a raw Desktop bridge outside the catalog SDK boundary ([pinned source](https://raw.githubusercontent.com/akinduroifedayo/living-spaces/e6145ccbd028edf55417822020426f4830343eb6/src/living-spaces/README.md))."
    },
    {
      "rank": 18,
      "pr": 153020,
      "name": "breadcrumbs",
      "total": 25,
      "why": "U4/R2/P4/C5/B5: The provenance-oriented memory design is creative, but the pinned summarizer subprocess has no explicit environment and therefore inherits the full process environment, potentially exposing provider secrets to that child, a catalog blocker despite the otherwise useful memory system ([pinned source](https://raw.githubusercontent.com/breadcrumbs-agency/hermes-breadcrumb-plugin/b8975645e5174edbcc00d5dc86a78e536bce938b/breadcrumbs/summarize.py))."
    },
    {
      "rank": 19,
      "pr": 153714,
      "name": "tenure",
      "total": 24,
      "why": "U4/R1/P5/C4/B5: Adaptive per-action trust and undo are useful, but the pinned HERMES_VAR matcher catastrophically backtracks on a short slash-run argument, reproduced locally, so a terminal hook can stall the process and reliability is 1 ([pinned source](https://raw.githubusercontent.com/otsoa-dubief/tenure/f4a62855e592b9e170ab0c54d9278aabef05f370/tenure/classify.py))."
    },
    {
      "rank": 20,
      "pr": 152651,
      "name": "workroom",
      "total": 23,
      "why": "U3/R1/P5/C4/B5: The document-release concept is valuable to a narrower audience, but the pinned uncapped _CD scan is quadratic on newline input even without a registered project, reproduced locally, leaving a process-stalling hook that later PR fixes cannot rescue at this pin ([pinned source](https://raw.githubusercontent.com/Pratiikpy/workroom-for-hermes/c5f6bf334c1163951e203dc7c5777faadfe49795/adapters/snapshotter.py))."
    }
  ]
}