revenantworks

Nine standalone wrights: skillwright (skills & packs, their eval suites, slims and trigger diagnosis), promptwright (prompts & model tiers), grillwright (the interview before any build), agentwright (agent systems), rigwright (standing Claude config), dispatchwright (fan-out dispatch), handoffwright (committed session handoffs), pacewright (usage pacing), scoutwright (Claude platform change watch). Each routes on its own description and works alone.

9 skills

Skills for running work on this machine, one program per runner. Seven members: lmstudiorunner (local LM Studio models, verified), comfyrunner (ComfyUI renders behind GPU guards), dockerrunner (Docker Desktop and WSL), duckrunner (DuckDB over a repo's data files), whisperrunner (local transcripts with whisper.cpp), hypervrunner (Hyper-V VMs and Windows Sandbox), obsrunner (OBS Studio, read-only by default, and FFmpeg post-processing). Each routes on its own description and works alone.

7 skills

Game-development skills on the -smith motif. Five members: pixelsmith (pixel art that reads at several zoom scales), godotsmith (Godot 4.x conventions and proof a build is green), soundsmith (game audio judged by numbers), slicesmith (the test-first coding loop), uismith (menus, HUD and dashboards with a fit probe at every width and theme). Standard-library scripts allowed when declared; Playwright is uismith's one optional package.

5 skills

Skills on the -scribe motif for words and knowledge. Five members: researchscribe (graded research verdicts, playbooks and reports), brandscribe (brand definitions, design systems and voice files), commscribe (messages and docs shaped to their channel, with a fact check), lorescribe (a fiction project's story bible and canon checks), vaultscribe (an LLM-maintained knowledge vault with lint and drift checks). Each routes on its own description and works alone.

5 skills

Skills that guard what Claude writes or manages. Four members: shieldwarden (leak and injection scans, naming and identity policy, names that advertise secrets, gated history rewrites that never push), trustwarden (pre-install vets of third-party code), keywarden (credential lifecycle, fingerprints only), gatewarden (runtime permission policy, owner-installed hooks, a settings map, and disk space and Claude's reach from evidence). Each routes on its own description and works alone.

4 skills

GDScript language server for Claude Code through Godot's own LSP: a stdio-to-TCP bridge (loopback only, Python stdlib) to the editor's port 6005, with an opt-in headless editor, plus a check-only GDScript format hook that reports NOT-RUN when no formatter is installed. Installs nothing.

hooks · LSP

Keeps brand definitions, several side by side, and turns one into a Claude Design System, DESIGN.md, VOICE.md or guide. Trigger to define a brand, palette, type or voice; turn a style guide or site into tokens; make, connect or sync a design system; audit for brand drift (off-palette colours, old names); pass or fail a page against a brand, or apply one to a README or generated output; critique or check a built UI, or a built site (dist/) for UI problems before it ships (hierarchy, spacing, states, generic look); or say /rw-brandscribe (setup, build, export, sync, audit, gate, ui). Microcopy and messages are commscribe's; React libraries design-sync's; charts dataviz's; canon lorescribe's; a pre-build brief, fit across widths and themes, and game UI uismith's; no logos.

1 skill

Finds the real cost of a trip and catches price drops after booking. Gives one all-in total per hotel, flight or rental car across sites, every tax, resort fee, city tax and parking added; pulls money-only complaints from recent reviews; checks a booked refundable stay or fare for a drop (rebook first, cancel second); puts flights, check-ins and cancel deadlines on the calendar from confirmations; checks a delayed or cancelled flight against EU261, UK261 and US DOT rules and drafts the claim. Trigger on compare these hotels, real total, hidden fees, fees reviewers mention, resort fee, watch my booking, did my hotel get cheaper, add my trip to my calendar, my flight was delayed, can I get a refund, or say faretender (compare, fees, watch, calendar, claim). Never books, cancels or pays. Points and miles are travel-hacking-toolkit's; the daily schedule agentwright's; vetting a connector trustwarden's.

1 skill

A small, security-first Godot MCP server for Claude Code: an editor plugin (loopback port, per-session token) and a Python stdlib stdio server. Eight tools: status, run and stop a scene, read game and editor output, scene tree, add a node and set a property (both undoable), save the open scene. Read-only switch on both sides; nothing deletes, writes other files or runs code sent to it. Installs nothing.

1 MCP server

Personal and household life skills. First member: faretender, the real cost of a trip and price drops after booking.

1 skill

Paces a Claude subscription's usage — how fast to spend this week, what can still run at N% weekly, whether planned work fits before a reset. Reads the 5-hour, weekly, per-model and CI-minutes meters, sets a spend mode (normal, pace, turbo, overnight, work-day, owner-away, reset-eve) and re-cuts waves that don't fit. Also when something ate usage overnight, CI minutes run low, context runs high (compact or clear?), a new model needs a measured baseline, or which skill or subagent used tokens; or say /rw-pacewright (check, mode, fit, context, uba, baseline, spend, refresh). The plan table is dispatchwright's; the handoff, handoffwright's; one artifact's cost, skillwright's slim; the meter statusline, rigwright's; a scheduled check, agentwright's.

1 skill

Directs pixel art that must read at several zoom scales at once — per-scale palette and silhouette rules, terrain-versus-unit contrast, a one-scene look test, and briefs for an artist or image generator. Trigger when a game renders one world through zoom bands or levels of detail, a pixelated 3D or 2.5D band with billboards included; when a sprite, unit or icon vanishes against terrain zoomed out; for a cross-band look test or pixel-art style guide; or say /rw-pixelsmith (test, brief, audit, 3d). Works from a described scene. Drawing is an art tool's job; rendering and LOD code the game's engineering; a brand palette brandscribe's; game audio soundsmith's.

1 skill

Vets third-party skills, plugins, MCP servers, GitHub Actions and model weights before install, with a verdict (PASS, CONDITIONAL, HOLD, FAIL) and install terms. Trigger on any should-I-install question, even with its files or hooks pasted in or a scan already clean; is this skill safe, vet this MCP server, check this GitHub Action, is this checkpoint safe or commercially usable, what does this installer write, a licence check before a fold-in, what changed in a plugin update; or say /rw-trustwarden (vet, revet, terms, refresh). Uses skillspector, zizmor, pinact and Scorecard when present; lists files no scanner read. Runtime permissions are gatewarden's; secrets shieldwarden's; your own skills skillwright's; sandboxing a download hypervrunner's.

1 skill

Transcribes local audio and video to txt, srt, vtt and json with timestamps using whisper.cpp, and proves which backend ran. Trigger to transcribe, caption or subtitle a recording, meeting, podcast or video locally; to pick a Whisper model for this GPU or estimate a batch; when a transcript is empty, cut short or repeating; when Whisper ran on the CPU or a pasted log must prove the GPU ran; when it shares the GPU with LM Studio or ComfyUI; or say /rw-whisperrunner (audit, check, plan, run, status, refresh). Never a cloud API or a download. Voice-overs are comfyrunner's; clips and burned-in captions obsrunner's; game audio soundsmith's; summarising transcripts lmstudiorunner's.

1 skill

Revenantworks shroud mod for Claude Code 2.1.287+: masks secrets and stream keys before Claude reads them (restoring a value only into the gitignored file it came from), flags a credential keywarden marked for rotation, and marks web, MCP and mail results and messages from routines or other sessions as data. Kept apart from the vigil plugin, so safety never depends on the dashboard. Never blocks your work; keeps no record; contacts no service.

Revenantworks dashboard mod for Claude Code 2.1.287+: which skills are enabled and how each fires (slash or automatic), failures, likely misroutes, tokens and cost by skill; the 5-hour, weekly, context and cache meters on the status line; health (mods loaded, hook drift, gatewarden's event counts); background tasks; optional research, GPU-lease, Godot and PR panels; and a suggestion queue that proposes, never builds. /rw-vigil opens it; Claude reads it only through vigil_read. Counts only, never text; counts stay in this session until /rw-vigil records on.