{
  "batch": "final-06",
  "ranking": [
    {
      "rank": 1,
      "pr": 149640,
      "name": "hermes-ark",
      "total": 29,
      "why": "Useful/Reliable/Polish/Creative/Beyond=5/5/5/4/5: The strongest everyday protection for an invested Hermes user combines automatic checkpoints, encrypted off-site backups and human-controlled item restore, with a finished Desktop and a tested restore engine ([pinned source](https://raw.githubusercontent.com/medomohamed22/hermes-ark/a23e529c1f636aa49d974576cdc5cb15fc8242d3/README.md))."
    },
    {
      "rank": 2,
      "pr": 150840,
      "name": "kata",
      "total": 29,
      "why": "Useful/Reliable/Polish/Creative/Beyond=5/4/5/5/5: Learning repeated successful tool sequences and compiling approved routines into deterministic tools most directly improves the agent itself, with excellent review UX and schema-drift checks, although timed-out worker threads are not forcibly stopped ([pinned source](https://raw.githubusercontent.com/opeblow/kata/cc59b08f1bbeeb02524c060faac04edb34de0564/README.md))."
    },
    {
      "rank": 3,
      "pr": 150353,
      "name": "fluent",
      "total": 28,
      "why": "Useful/Reliable/Polish/Creative/Beyond=5/5/5/3/5: Checking invented flags against installed help before execution is a broadly useful daily improvement with conservative confidence checks and normal Hermes approvals, while remote terminal backends deliberately receive no enforcement ([pinned source](https://raw.githubusercontent.com/Bruno-coder21/fluent/182c4301f2437e784e9022f1f38576e1d68c3f5f/README.md))."
    },
    {
      "rank": 4,
      "pr": 153844,
      "name": "fine-print",
      "total": 28,
      "why": "Useful/Reliable/Polish/Creative/Beyond=5/4/5/4/5: Making truncated results, missing images and provider fallbacks visible improves almost every agent workflow with restrained actionable notices, but version-specific result detectors and fixture-only coverage keep reliability below the best infrastructure entries ([pinned source](https://raw.githubusercontent.com/Joko-F/hermes-fine-print/7b574f1b5c8dc8eafe3ea9f87ae6b0111853f0ba/README.md))."
    },
    {
      "rank": 5,
      "pr": 152458,
      "name": "housemate",
      "total": 28,
      "why": "Useful/Reliable/Polish/Creative/Beyond=4/4/5/5/5: Turning spare phones into paired room speakers with arbitration, persistent timers and native approval cards is an unusually complete extension of Hermes onto personal hardware, with more setup and audio-device uncertainty than the leading install-and-use utilities ([pinned source](https://raw.githubusercontent.com/RaYYeR220/housemate/09c1a1d0a88bfa8762cee6b3ccf94271b0f172c5/README.md))."
    },
    {
      "rank": 6,
      "pr": 150596,
      "name": "ariadne",
      "total": 28,
      "why": "Useful/Reliable/Polish/Creative/Beyond=4/5/5/4/5: Cassette-backed replay through model and tool middleware gives Hermes users a polished offline regression and upgrade-checking capability a skill cannot supply, but its developer-oriented use is less continuous than the higher daily-use entries ([pinned source](https://raw.githubusercontent.com/miracle352/hermes-ariadne/128c95967a0761ad24d3d224441d1502c43bbb69/README.md))."
    },
    {
      "rank": 7,
      "pr": 150556,
      "name": "hermes-ui-cards",
      "total": 28,
      "why": "Useful/Reliable/Polish/Creative/Beyond=5/4/5/4/5: Validated interactive cards with persistent drafts, text fallbacks and real chat replies improve many Desktop interactions, but concurrent-client answers are only eventually deduplicated and confirmation cards remain messages rather than enforced execution approvals ([pinned source](https://raw.githubusercontent.com/robbyczgw-cla/hermes-ui-cards/c40df4d41cd5439cf4902995fd4a9e5505483f25/README.md))."
    },
    {
      "rank": 8,
      "pr": 151887,
      "name": "cmi",
      "total": 27,
      "why": "Useful/Reliable/Polish/Creative/Beyond=5/4/4/4/5: The native context engine reduces repeated context while preserving recallable originals and tool pairs, making it highly retainable, but internal Hermes dependencies and less polished native-plugin packaging trail the leaders and the separately installed Console is not credited ([pinned source](https://raw.githubusercontent.com/jeremiebergel/context-manager-intelligence/399468f0687c051476961e4bb045eb84263be6b1/docs/REFERENCE.md))."
    },
    {
      "rank": 9,
      "pr": 150800,
      "name": "hypnos",
      "total": 27,
      "why": "Useful/Reliable/Polish/Creative/Beyond=3/4/5/5/5: The most ambitious model-level extension implements personal LoRA practice, held-out promotion and fallback routing rather than merely describing learning, but a separate trainer and narrow hardware qualification limit day-one adoption and full training was not exercised here ([pinned source](https://raw.githubusercontent.com/0xNoramiya/hypnos/0a809b9085fdeaed63fd909b927afcc41de512de/README.md))."
    },
    {
      "rank": 10,
      "pr": 152691,
      "name": "grimoire",
      "total": 27,
      "why": "Useful/Reliable/Polish/Creative/Beyond=4/5/5/3/5: A constrained read-only DuckDB engine plus query receipts and numeric-answer checks makes local data analysis dependable and polished, although its familiar analytics design and data-file audience keep it below the broader agent improvements ([pinned source](https://raw.githubusercontent.com/bettersandeep/grimoire/4824773040b131f3179e5896d71d1435908758a7/engine.py))."
    },
    {
      "rank": 11,
      "pr": 150644,
      "name": "hermes-canvas",
      "total": 27,
      "why": "Useful/Reliable/Polish/Creative/Beyond=4/5/5/3/5: A polished plan canvas tied to real writes, exact check exit codes and user vetoes makes coding work steerable through runtime hooks, but shell-write recognition is best effort and the benefit is concentrated in multi-step Desktop coding ([pinned source](https://raw.githubusercontent.com/0xbyt4/hermes-canvas/d6eeff0d6f7ec653262537bdd7bb001cc6d3c7f8/README.md))."
    },
    {
      "rank": 12,
      "pr": 152908,
      "name": "contract-necromancer",
      "total": 27,
      "why": "Useful/Reliable/Polish/Creative/Beyond=4/5/4/4/5: Human-frozen tests, tamper detection and independently executed completion checks provide stronger evidence of done than a prompt or ordinary check-runner, with clear receipts and bounded execution but a narrower coding workflow and less visual polish ([pinned source](https://raw.githubusercontent.com/vijaygopalbalasa/hermes-necromancer/0ff7bec40aba10091a84f94025d4471984fd7386/README.md))."
    },
    {
      "rank": 13,
      "pr": 153851,
      "name": "foreman",
      "total": 27,
      "why": "Useful/Reliable/Polish/Creative/Beyond=4/4/5/4/5: Inspecting real diffs and base-branch check configuration before publishing catches test-rigging shortcuts with a polished human review pipeline, but hidden command routes and approval-dependent incomplete inspections make it weaker enforcement than frozen contracts ([pinned source](https://raw.githubusercontent.com/PedValme/foreman/c15453cb2fc07fa674d6f7b548a553b3ce9e6f5d/plugin/README.md))."
    },
    {
      "rank": 14,
      "pr": 153955,
      "name": "shadow",
      "total": 27,
      "why": "Useful/Reliable/Polish/Creative/Beyond=4/4/5/4/5: Blind two-order comparisons on saved conversations, isolation verification and undoable model switching help users improve their own agent, but provider spending, substantial private-API coupling and approximate replays make this an occasional tuning tool rather than a universal daily install ([pinned source](https://raw.githubusercontent.com/akinduroifedayo/shadow/38173bc9c696df3b1f87b5b9b60ccecd36a260ab/README.md))."
    },
    {
      "rank": 15,
      "pr": 149738,
      "name": "trust-ladder",
      "total": 26,
      "why": "Useful/Reliable/Polish/Creative/Beyond=5/3/4/4/5: Persistent correction guards and verified-task gates have excellent broad utility, but at this pin the profiler accepts ^(a|aa)+$ and a 36-a near-miss exceeds a three-second isolated matching check, undermining its no-stall claim ([pinned source](https://raw.githubusercontent.com/merdoks/trust-ladder/fae4e9df58767e5e3832c8f65936b97868bc98fc/compiler.py))."
    },
    {
      "rank": 16,
      "pr": 150251,
      "name": "gemini-live",
      "total": 26,
      "why": "Useful/Reliable/Polish/Creative/Beyond=4/4/5/3/5: Full-duplex Desktop voice with real-agent delegation, server-held credentials and person-only approval handling is product-like and useful, but a Gemini account and a familiar voice-integration concept place it behind the stronger agent-learning and verification entries and live calls were not exercised here ([pinned source](https://raw.githubusercontent.com/cineglobe/hermes-gemini-live/b6c2f24354aeed1a8b5651b62944d14ed96a472d/README.md))."
    },
    {
      "rank": 17,
      "pr": 153809,
      "name": "living-spaces",
      "total": 26,
      "why": "Useful/Reliable/Polish/Creative/Beyond=4/3/5/4/5: Persistent sandboxed apps with shared records, revisions and native upkeep substantially expand Desktop, but documented WebRTC egress, plaintext Space credentials and private cron/raw-bridge integration leave a larger trust and compatibility burden than the higher-ranked everyday tools ([pinned source](https://raw.githubusercontent.com/akinduroifedayo/living-spaces/e6145ccbd028edf55417822020426f4830343eb6/src/living-spaces/README.md))."
    },
    {
      "rank": 18,
      "pr": 153020,
      "name": "breadcrumbs",
      "total": 25,
      "why": "Useful/Reliable/Polish/Creative/Beyond=4/2/4/5/5: Addressable verbatim memory, adjustable recall and a context engine are unusually creative core-agent improvements, but the pinned summarizer launches llama-completion without an explicit environment and therefore inherits provider credentials when that optional path is active ([pinned source](https://raw.githubusercontent.com/breadcrumbs-agency/hermes-breadcrumb-plugin/b8975645e5174edbcc00d5dc86a78e536bce938b/breadcrumbs/summarize.py))."
    },
    {
      "rank": 19,
      "pr": 152651,
      "name": "workroom",
      "total": 25,
      "why": "Useful/Reliable/Polish/Creative/Beyond=4/2/5/4/5: Source-linked Office edits, staged human releases and a thorough Desktop are valuable, but the assigned pin still has quadratic snapshotter and HTML-stripping regex paths plus overly broad project registration, and isolated probes confirmed the newline slowdown despite later fixes outside this pin ([pinned source](https://raw.githubusercontent.com/Pratiikpy/workroom-for-hermes/c5f6bf334c1163951e203dc7c5777faadfe49795/adapters/snapshotter.py))."
    },
    {
      "rank": 20,
      "pr": 153714,
      "name": "tenure",
      "total": 24,
      "why": "Useful/Reliable/Polish/Creative/Beyond=4/1/5/4/5: Revocable per-action trust, undo and an autonomy passport are thoughtfully polished, but the pinned terminal classifier has catastrophic HERMES_VAR backtracking on a short slash-run near-miss, making the agent-wide hook unsafe despite its passing admission check ([pinned source](https://raw.githubusercontent.com/otsoa-dubief/tenure/f4a62855e592b9e170ab0c54d9278aabef05f370/tenure/classify.py))."
    }
  ]
}