{ "batch": "read2-015", "plugins": [ { "pr": 0, "name": "done-bell", "does": "Notifies when Hermes finishes a turn or needs approval, with a Desktop status chip, chime and mute controls.", "useful": 4, "reliable": 3, "polish": 2, "creative": 2, "beyond": 4, "total": 19, "best": "A small, directly useful agent lifecycle integration works without external accounts.", "worst": "The README installs from a developer-specific Windows path and requires manual Desktop copying; tests cover Python state and hooks but not the Desktop notification path.", "standout": false, "blocked": null }, { "pr": 0, "name": "homeassistant", "does": "Connects Hermes to Home Assistant device tools, filtered state-change events and notification delivery.", "useful": 4, "reliable": 4, "polish": 4, "creative": 3, "beyond": 5, "total": 25, "best": "A real gateway platform combines reactive home events with device control, scoped secrets and tested entity restrictions.", "worst": "It requires a Home Assistant installation and careful operator filtering; unrestricted service calls otherwise permit broad device control.", "standout": false, "blocked": null }, { "pr": 153976, "name": "model-kombat", "does": "Runs two Hermes models on mechanically scored task cards and records lane ratings, costs and Desktop/kanban scoreboards.", "useful": 4, "reliable": 3, "polish": 5, "creative": 4, "beyond": 4, "total": 24, "best": "Deterministic checks, honest draws, usage accounting and a substantial visual interface turn model comparisons into usable evidence.", "worst": "The scoring tool accepts allow_code_execution from tool arguments over the disabled configuration default, while execution remains only a best-effort Python subprocess boundary.", "standout": false, "blocked": null }, { "pr": 149647, "name": "solverforge-planner", "does": "Fits tasks around meetings and constraints using a Rust solver, with approved local calendar application and cloud-calendar handoff.", "useful": 4, "reliable": 5, "polish": 5, "creative": 4, "beyond": 4, "total": 26, "best": "Freshness checks, proposal fingerprints, scoped approvals and real-engine/calendar tests support an unusually complete planning workflow.", "worst": "Cloud calendar reads and writes depend on Hermes coordinating a separate calendar tool, and first use requires a platform-specific engine download.", "standout": true, "blocked": null }, { "pr": 152842, "name": "speckle", "does": "Lets Hermes query, compare and watch Speckle BIM models and manipulate a bundled Desktop 3D viewer.", "useful": 3, "reliable": 3, "polish": 5, "creative": 4, "beyond": 4, "total": 23, "best": "Model selection, element queries, visual diffs and saved version checks form a coherent specialist workflow beyond chat.", "worst": "Issue creation immediately writes through the API; asking the user first is only a prompt/schema instruction, and tests focus on indexing, checks and redirects rather than the full viewer/watcher path.", "standout": false, "blocked": null }, { "pr": 149503, "name": "elcaro", "does": "Scans tool results and arguments for prompt injection, quarantines detections, gates tainted-session actions and records a tamper-evident threat log.", "useful": 5, "reliable": 3, "polish": 4, "creative": 4, "beyond": 5, "total": 26, "best": "Result replacement, inherited session taint and action-sequence policies protect the agent at boundaries a skill cannot enforce.", "worst": "The default guard scans only the first 48,000 characters, skips error envelopes and fails open on scan errors, so the README promise that injection never reaches the model is too absolute.", "standout": false, "blocked": null }, { "pr": 150453, "name": "workgraph", "does": "Captures agent work as branching source/data snapshots with browser evidence, recoverable workspaces and operator-controlled acceptance.", "useful": 4, "reliable": 4, "polish": 4, "creative": 5, "beyond": 5, "total": 27, "best": "Native tool middleware checkpoints edits and avoids repeating uncertain side effects, while the visual history preserves executable product states.", "worst": "It is a Linux developer preview requiring Git, bubblewrap and Chromium for its full workflow, and only supported foreground operations receive automatic capture.", "standout": true, "blocked": null }, { "pr": 152504, "name": "benchforge", "does": "Compiles Arduino Uno firmware, simulates chip-level behavior and fault cases, and gates recognized flash commands on approved passing receipts.", "useful": 3, "reliable": 4, "polish": 5, "creative": 5, "beyond": 5, "total": 27, "best": "Human-approved checks, deterministic sensor-fault simulation and a bounded repair loop give Hermes concrete hardware-testing capability with inspectable evidence.", "worst": "Coverage is limited to the Uno and idealized supported parts, and the flash gate recognizes terminal command patterns rather than enforcing a universal hardware-access boundary.", "standout": true, "blocked": null }, { "pr": 151371, "name": "smart-dictation", "does": "Adds cached local speech recognition and optional model correction to native Hermes dictation, plus a contextual Desktop review workflow.", "useful": 4, "reliable": 4, "polish": 4, "creative": 3, "beyond": 5, "total": 25, "best": "Profile-bound decoding, explicit correction consent and raw-transcript recovery pair a real transcription provider with careful draft-placement safeguards.", "worst": "Setup is per profile and requires pre-cached model weights; native mode inherits host insertion behavior and cannot reliably interrupt an active decoder.", "standout": false, "blocked": null }, { "pr": 153737, "name": "backfill", "does": "Replays historical Hermes conversations into the currently configured memory provider with dry runs, resume tracking and recall checks.", "useful": 4, "reliable": 3, "polish": 3, "creative": 4, "beyond": 5, "total": 24, "best": "A provider-independent migration path preserves the value of old chats while using Hermes memory redaction and ordinary provider hooks.", "worst": "Original turn timestamps are lost, cloud providers remain untested, and resume bookkeeping is keyed by provider name rather than a particular destination store.", "standout": false, "blocked": null }, { "pr": 150832, "name": "cacheguard", "does": "Warns or blocks large sends after idle periods and reports measured prompt-cache reuse through Hermes request and execution middleware.", "useful": 4, "reliable": 4, "polish": 4, "creative": 4, "beyond": 5, "total": 26, "best": "The send guard, retry handling and usage journal improve the user’s own agent without extra model calls or stored message text.", "worst": "Cache expiry cannot be known before sending, and cache accounting and unattended-mode detection depend on private Hermes helpers that may degrade across releases.", "standout": false, "blocked": null }, { "pr": 153020, "name": "breadcrumbs", "does": "Stores exact conversation crumbs for adjustable-detail recall and replaces long-context compaction with a retrievable crumb layer and provenance explorer.", "useful": 4, "reliable": 4, "polish": 5, "creative": 5, "beyond": 5, "total": 28, "best": "Memory tools and a genuine context engine share exact-source addresses and receipts, allowing folded history to be reopened without model-written summaries.", "worst": "CRC32C collisions use local ordinal suffixes rather than globally unique content hashes, while Windows lacks the multi-writer file locking used by the store on Unix.", "standout": true, "blocked": null } ] }