{ "batch": "deep-13", "setup": "Cloned Hermes source at fdde27e7dae174cb4ba842589e92508852b7dd9f and installed it with uv sync --locked --python 3.14. Each pinned plugin passed through tryit.py with Nous FakeLLMServer and a real AIAgent conversation; empty plans showed Hermes tool registration and scripted plans exercised calls. Plugin tools were deferred behind Hermes tool_search/tool_call, so the first model offer listed none directly.", "plugins": [ { "pr": 153826, "name": "googlehome", "validate": "Passed all catalog checks; warnings flag private Hermes core imports and two raw Desktop bridge calls for manual review.", "inside": "Loaded as a platform; doctor registered five tools and two hooks. Four scripted calls ran: status clearly reported no Cast device, discovery returned an empty list, an invalid volume was rejected by Hermes schema validation, and empty speech returned a text-required error.", "tests": "293 passed, 1 skipped (294 total) in Python unittest; 37/37 Desktop checks passed.", "useful": 4, "reliable": 4, "polish": 5, "creative": 4, "beyond": 5, "total": 27, "rank": 5, "best": "The Cast platform, speech tools, and Desktop controls make a local speaker a Hermes voice endpoint, with broad mock-device and UI test coverage.", "worst": "A physical speaker and usable TTS backend were absent here, while private core imports and raw Desktop bridge calls leave compatibility risk.", "blocked": null }, { "pr": 151135, "name": "wake-me-when", "validate": "Passed every catalog check with no warnings.", "inside": "Loaded; doctor registered seven tools and one hook. Five calls ran: an empty watch list, a bundled demo watch started, the active watch appeared, stop-all ended it, and an unknown watch ID returned a clear message. The scripted turn stopped the watch before an alert.", "tests": "115 passed, 7 skipped in an isolated test environment; an initial Hermes-environment run had shared-storage interference and was superseded by this clean run.", "useful": 4, "reliable": 4, "polish": 5, "creative": 5, "beyond": 5, "total": 28, "rank": 4, "best": "It turns a live stream into event-driven agent attention with transcript evidence, clips, catch-up, and feedback rather than a one-shot search.", "worst": "Continuous live capture needs external media tools and speech setup; this scripted conversation verified watch control but not a sustained real-stream alert.", "blocked": null }, { "pr": 150110, "name": "hermes-colab", "validate": "Passed every catalog check with no warnings.", "inside": "Loaded; doctor registered eight tools and one hook. Five calls ran: doctor identified missing Google OAuth client values, runtimes gave the sign-in step, recent jobs returned an empty list, an invalid upload path was refused, and an unknown log ID returned no such run.", "tests": "61 passed, 5 skipped in Python pytest.", "useful": 3, "reliable": 4, "polish": 5, "creative": 4, "beyond": 5, "total": 26, "rank": 6, "best": "Background Colab jobs, output syncing, resume logic, and a Desktop run view connect Hermes chat to useful remote compute.", "worst": "The actual GPU execution and recovery path requires a user-owned Google OAuth client and account, so it could only be judged from code and local tests here.", "blocked": null }, { "pr": 153596, "name": "sentinel-vision", "validate": "Passed every catalog check with no warnings.", "inside": "Loaded; doctor registered nine tools and a slash command. Five calls ran: source listing and setup check identified the absent display/camera and alert opt-in, rules were empty, a camera watch rule was created with a fallback compiler, and a malformed time filter returned a clear error; no frame or alert was observed.", "tests": "1,679 passed, 2 skipped in Python pytest.", "useful": 4, "reliable": 5, "polish": 5, "creative": 4, "beyond": 5, "total": 28, "rank": 2, "best": "Persistent visual rules, local change filtering, model-call budgets, and a Desktop pane make Hermes able to watch a real scene and report evidence.", "worst": "This headless run lacked a camera or display and vision model, so rule creation did not prove frame judging or alert delivery end to end.", "blocked": null }, { "pr": 152481, "name": "argus", "validate": "Passed all catalog checks; a warning identifies private Hermes imports in skill-apply paths.", "inside": "Loaded; doctor registered five hooks, an auxiliary task, a slash command, and a CLI command, with no agent tool. Three Hermes terminal calls ran: status reported no prior plugin data, backfill completed with zero tool outcomes and skill edits, and applying an unknown proposal was refused without an accept decision.", "tests": "115 passed in Python pytest after installing the declared mypy test dependency.", "useful": 5, "reliable": 4, "polish": 5, "creative": 5, "beyond": 5, "total": 29, "rank": 1, "best": "Intent-based review and approval-gated skill changes improve the user’s own Hermes agent across sessions, directly matching the hackathon’s agent-improvement goal.", "worst": "Deriving intent and reviews can consume many model calls, and private core imports in the apply path may break on Hermes upgrades.", "blocked": null }, { "pr": 153357, "name": "twring", "validate": "Passed every catalog check with no warnings.", "inside": "Loaded as a platform; doctor registered its adapter and CLI, with no agent tool. Three Hermes terminal calls ran: status said unlinked, linking to a non-HTTPS server failed immediately, and unlink completed cleanly. A real call could not run without the app and account.", "tests": "84 passed in the Hermes integration Python pytest suite.", "useful": 5, "reliable": 4, "polish": 5, "creative": 4, "beyond": 5, "total": 28, "rank": 3, "best": "Phone calls, shared voice and chat turns, approval cards, and scheduled-job rings make a personal Hermes agent reachable throughout the day.", "worst": "The main phone experience depends on a Twring app and linked account, so the live call path was not independently exercised here.", "blocked": null } ] }