{ "batch": "read2-005", "plugins": [ { "pr": 0, "name": "done-bell", "does": "Rings on completed turns and approval requests, with CLI notifications, a Desktop status chip and mute controls.", "useful": 4, "reliable": 3, "polish": 3, "creative": 2, "beyond": 4, "total": 20, "best": "Actual lifecycle hooks and a Desktop busy-state watcher make leaving a long agent task practical without another service.", "worst": "Setup uses a developer-specific Windows path and manual pane copying, while tests suppress OS notifications and do not exercise the Desktop watcher.", "standout": false, "blocked": null }, { "pr": 0, "name": "homeassistant", "does": "Connects Hermes to Home Assistant state-change events and provides entity discovery, state reads, service calls and notification delivery.", "useful": 4, "reliable": 4, "polish": 4, "creative": 3, "beyond": 5, "total": 25, "best": "A genuine gateway platform combines filtered, cooled-down inbound events with scoped secrets, service-domain restrictions and extensive policy tests.", "worst": "Device-control entity restrictions are opt-in, and the broad token plus permitted scripts still require careful operator configuration.", "standout": false, "blocked": null }, { "pr": 153343, "name": "worktree-sandbox", "does": "Routes terminal and file work into Git worktrees, races candidate edits against verification commands and promotes reviewed changes through tools or Desktop.", "useful": 4, "reliable": 2, "polish": 4, "creative": 4, "beyond": 5, "total": 24, "best": "A real environment provider, diff UI, backups and deterministic candidate scoring improve the agent runtime beyond a skill.", "worst": "Promotion writes target paths without resolving destination symlinks, and model-supplied race shell commands execute directly without a visible approval gate.", "standout": false, "blocked": null }, { "pr": 150293, "name": "procgen-studio", "does": "Adds image and video providers that turn the configured model's drawing programs into local PNG/MP4 assets with editable source, seeds and style contracts.", "useful": 4, "reliable": 4, "polish": 5, "creative": 4, "beyond": 5, "total": 27, "best": "Provider integration, local rerolls/restyles, a bounded critic loop, Desktop gallery and adversarial render tests form a coherent creative product.", "worst": "Drawing quality depends heavily on the coding model, and executing generated Python relies on language-level guards rather than an OS security boundary.", "standout": true, "blocked": null }, { "pr": 152248, "name": "rainbowomma", "does": "Gives Hermes profiles on a shared Mac separate browsers and resource leases, blocks common cross-agent collisions and cleans up owned resources.", "useful": 4, "reliable": 4, "polish": 4, "creative": 4, "beyond": 5, "total": 26, "best": "A browser provider plus fail-closed pre-tool guards and ownership-aware process/document cleanup directly improve agents running on the user's own hardware.", "worst": "The guard inspects command text rather than confining programs, and macOS automation permissions and cooperative profile naming limit its guarantees.", "standout": false, "blocked": null }, { "pr": 152408, "name": "rich-paste", "does": "Converts Desktop clipboard content into structured Markdown and interleaves attached images at their text labels in outgoing model requests.", "useful": 5, "reliable": 4, "polish": 5, "creative": 4, "beyond": 5, "total": 28, "best": "Composer integration and request middleware solve image-to-paragraph binding across message formats, backed by extensive conversion, request-shape and safety tests.", "worst": "Screenshot binding falls back to attachment order after a reload or for large images, and cache expiry can break image display in old chats.", "standout": true, "blocked": null }, { "pr": 153531, "name": "hermes-reliability", "does": "Runs source-grounded agent-testing campaigns with generated scenarios, independent reviews, execution evidence and frozen regression benchmarks.", "useful": 3, "reliable": 2, "polish": 4, "creative": 4, "beyond": 4, "total": 21, "best": "Immutable benchmark provenance and validity-aware comparisons provide a serious regression workflow with native tools, background workers and a dashboard.", "worst": "At the field pin evaluator preflight inherits the full environment and execution approval uses a global rule key; the documented complete autonomous live campaign also remains unverified.", "standout": false, "blocked": null }, { "pr": 151252, "name": "proof-run", "does": "Records demonstrated tool actions and gates later covered mutations on exact arguments, fresh session-bound observations and matching result evidence.", "useful": 3, "reliable": 4, "polish": 3, "creative": 4, "beyond": 5, "total": 24, "best": "Consumable read observations, durable unproven states and carefully tested atomic storage make a concrete fail-closed tool-boundary guard.", "worst": "It permits only exact demonstrated actions, cannot enforce workflow ordering or stop alternate tool routes, and cannot undo a mutation whose result fails its evidence check.", "standout": false, "blocked": null }, { "pr": 149856, "name": "fledge", "does": "Holds skills learned from visibly bad runs for review, puts saved skills on probation and protects hand-edited skills from autonomous rewrites.", "useful": 5, "reliable": 4, "polish": 4, "creative": 4, "beyond": 5, "total": 27, "best": "Nine real lifecycle hooks govern Hermes's self-improvement loop using run evidence, human review and recoverable skill probation without another model.", "worst": "Heuristic correction/error detection misses silent wrong answers, and private queue/curator imports plus deliberate fail-open behavior weaken protection when integration fails.", "standout": true, "blocked": null }, { "pr": 151158, "name": "skill-layers", "does": "Preserves installed skill sources, creates writable derivatives and prepares reviewed three-way updates or rollback copies with new native skill names.", "useful": 4, "reliable": 5, "polish": 4, "creative": 3, "beyond": 4, "total": 24, "best": "Revision and review-hash checks, immutable snapshots, no-replace publication and fault/path-boundary tests make source-preserving customization unusually careful.", "worst": "It does not switch the running session to the preferred derivative, cannot accept plugin-bundled sources, and textual merges still need human behavioral review.", "standout": false, "blocked": null }, { "pr": 152190, "name": "hermes-tuneup", "does": "Explains compaction triggers and wasteful settings across Hermes profiles and model routes, with read-only reports and explicitly confirmed configuration fixes.", "useful": 5, "reliable": 4, "polish": 4, "creative": 3, "beyond": 5, "total": 26, "best": "It uses Hermes's actual resolver, previews changes on a copy, backs up confirmed fixes and records optional content-free compaction receipts.", "worst": "Accuracy depends on undocumented core resolver APIs, and undeclared rarely used models can make a useful threshold rule appear dead.", "standout": false, "blocked": null }, { "pr": 151270, "name": "hermes-line", "does": "Adds an approval-gated Twilio phone service with menus, hold handling, callbacks, owner bridging, a Desktop pane and separately approved Android SMS-code delivery.", "useful": 4, "reliable": 3, "polish": 4, "creative": 4, "beyond": 5, "total": 25, "best": "Scoped single-use consent, restricted per-call Hermes sessions and extensive call/OTP tests support a real phone capability on the user's own account.", "worst": "Twilio, a public tunnel and a separate call profile make setup substantial, while live detection of a real agent and successful bridging remain unproven in the documented trials.", "standout": false, "blocked": null } ] }