Plugins by apoapostolov
16 plugins by apoapostolov in the catalog, most-starred first.
apoapostolovbetter-capabilitiesDelete and package skills, browse their markdown files, and save per-profile on/off presets from one Settings page. Disclosure — deleting a skill from its page moves the skill folder out of `<HERMES_HOME>/skills` (on Windows to the Recycle Bin by running `powershell.exe`, elsewhere to `<HERMES_HOME>/better-capabilities-trash/`); applying a preset switches skills, toolsets and agent plugins on or off, including for other profiles the preset names.CommunityDesktop app50apoapostolovbetter-session-appearanceSet a color for each session and show an idle glyph to make rows easier to scan.CommunityDesktop app50apoapostolovcdp-managerLocal Chrome CDP debug ports in the status bar (probe, launch, stop, recheck) plus a managed port Hermes keeps alive and the `cdp` agent tool, which also detects and repairs a Chrome profile-picker launch (wedged/unwedge). Disclosure: launches local Chrome on 127.0.0.1 only, with per-port profiles under the local app-data directory, and closes it with a clean Browser.close; state lives in the plugin's own config file. Windows only.CommunityDesktop app50apoapostolovdrag-to-pin-sessionPin or unpin sessions from their row with supported desktop SDK actions.CommunityDesktop app50apoapostolovhermes-jev-routingPer-turn model routing for the request that is about to go out. Jev reads the turn, picks the tier it deserves, and the plugin rewrites the provider request onto that model, including the reasoning level when the request carries one. Shadow by default, with notify, confirm and auto modes, a strip under the composer for the turn that was just routed, and a spend ledger. A pick on another provider is reported rather than sent, because a plugin cannot open a second provider client. Disclosure — once a routing JSON is configured, every turn (shadow mode included) sends the current message, up to 4,000 characters of recent user/assistant turns, the active model name, an estimated context size and the spend/budget snapshot to TypeSafe (api.typesafe.ai, endpoint overridable in that JSON), and the jev_route tool and /jev-routing route send their text the same way; in auto mode it rewrites the model and reasoning level on outgoing requests, and in Hermes Desktop the auto mode and the approve row also switch the session's model, including onto another provider (core's expensive-model confirmation is shown to the user, never answered automatically).Communitymodels50apoapostolovintelligent-tool-breakAbort a stalled in-flight tool call with /break, send a note with /break {message}, or retry it with /again without cancelling the rest of the task.CommunityDesktop app50apoapostoloviteration-budget-meterStatus-bar chip for the focused session per-turn iteration usage (N/60), with a hover tooltip and a click popover of per-request stats.CommunityDesktop app50apoapostolovmemory-reviewCheckbox staged memory writes and approve or reject the selection from a dialog.CommunityDesktop app50apoapostolovopaque-composerKeep the desktop composer solid while scrolling so conversation text stays readable.CommunityDesktop app50apoapostolovprompt-enhanceRewrite a composer draft outside the chat and put the finished prompt back in the field. A prompt library keeps saved prompts, and a one-call model pick does not change the session.CommunityDesktop app50apoapostolovprovider-statusStatus-bar quota tracking for 17 providers: Tavily, OpenCode Go, DeepSeek, GLM (z.ai), OpenRouter, Grok (xAI), Codex (OpenAI), OpenAI, Anthropic, Groq, Cerebras, Moonshot Kimi, MiniMax, Google Gemini, Hugging Face, Mistral, and Qwen. DeepSeek and GLM show peak/off-peak gauges and countdowns. Includes OAuth and multi-key pools. This edition does not rotate Hermes environment keys on exhaust, and an expired OAuth token asks for a new login. Codex and Grok checks use their CLI client identity and user-agent. Disclosure — Reads provider API keys from the Hermes .env and its own config.json and calls each enabled provider's own usage or billing endpoint; Codex and Grok sign-ins use those CLIs' OAuth client identity and user-agent, keep the plugin's tokens in its install folder, and ask for a new login on expiry. When the Discord gateway runs, the bot's Discord presence (visible to everyone in servers the bot shares) is updated every 20 seconds by default with tokens/sec, context-window use and the active provider's name with its remaining quota or dollar credit balance (model and running tool are opt-in); turn it off with plugins.entries.provider-status.settings.discord_presence_enabled: false. /pulse and the agent_pulse tool report process-wide telemetry (model, provider, current tool, quota) to anyone allowed to call them on a gateway. Writes a secret-free quota-cache.json into its install folder and reads Hermes config.yaml for the active model.CommunityDesktop app50apoapostolovreasoning-switchRotate the focused session reasoning effort from the status bar. A dialog sets the levels, colors, and per-level prompt limits. Disclosure — changes the focused session's reasoning effort through the gateway config.set API (core may fall back to writing the global config.yaml when the session id is stale) and auto-demotes the level after the configured number of prompts.CommunityDesktop app50apoapostolovrss-readerDesktop RSS reader with reader-mode capture, drag-reorder subscriptions, mute rules, saved searches, AI summaries, keyboard shortcuts, and an optional headline ticker. Disclosure — optional YouTube cookies you paste in Settings are sent with YouTube feed requests; selectable User-Agent presets (incl. a ChatGPT-User string) are sent to feed hosts; YouTube videos render as youtube-nocookie.com embed iframes inside the reader.CommunityDesktop app50apoapostolovsidebar-managerHide and reorder core sidebar navigation rows from a statusbar dialog.CommunityDesktop app50apoapostolovvercel-agent-desktop-supportVercel Agent Desktop Support: operate Windows apps through the accessibility tree instead of the screen. Reads a window's real UI as roles, names, states and refs, then clicks, fills a field, selects an option, toggles a checkbox, scrolls, sends a key combo, and manages windows and the clipboard. Actions are semantic and headless by default, so they do not take focus, move the cursor or synthesize keystrokes, and they fail closed rather than guess. desktop_install downloads the official vercel-labs/agent-desktop release for the running Windows target, verifies the archive against a sha256 pinned in the plugin, and proves the install by running version and status. Disclosure — the plugin runs the agent-desktop executable as a child process with arguments it builds, using the copy it installed or one already on PATH, in the npm global folder, AGENT_DESKTOP_BIN or the bin_path setting (those are not checksum-checked); starting or closing a program, installing the binary, and the first call that auto-installs a missing binary (auto_install, off by default) go through Hermes's approval prompt (Hermes Agent 0.18.1 or newer), while key presses, field values and clipboard writes into any window, a terminal included, run without a prompt; desktop_install is the only tool that uses the network, downloading the official vercel-labs/agent-desktop release from GitHub, and a release without a digest pinned in the plugin is refused; it reads window titles, labels, field values and the clipboard on request, which can contain private content. Windows only.Communityautomation50apoapostolovvoice-gateReviews changed prose with 75 mechanical rule families and optional contextual review against the complete humanizer skill and its references. Settings keep everyday controls visible and collapse Vale, bots, and individual rules. Slow inventory requests allow up to two minutes and offer retry; the page includes five-row exceptions and Vale installation guidance. Revision requests allow at most two opportunities and compare protected spans and quantities. Prose em dashes are flagged while bullet label separators and empty table values remain allowed. Reads the changed prose paths reported by Hermes; manual tools can check supplied text. Model review is off by default and sends reviewed documents to the configured Hermes provider when enabled. It provides editorial judgments, not authorship proof. Vale is optional; status, check, and install request approval before starting a program, turn-time findings use only a binary approved that way, and Vale runs with a short environment rather than the host's. No project-file writes or core overrides. Simple PDF text extraction has no OCR. Disclosure — reads the changed prose files Hermes reports after each turn and any text or paths given to its tools; optional model review (off by default) sends reviewed documents to the configured Hermes provider; optional Vale support downloads a sha256-pinned Vale release from GitHub into <HERMES_HOME>/plugin-data/voice-gate/ only after approval, and runs Vale (status, check, install, and turn-time advisory checks on changed files) only for a binary approved through Hermes' approval prompt and recorded by digest, with a minimal environment.Communitytools50
GitHub profile














