Make NPCs Talk

Before you start: read this whole install prompt, tell me in a few lines what this skill does and what installing it involves, and wait for my go-ahead before you do anything else. Prerequisites — paid external services are required to USE this skill, even though copying its files needs none: the workflow it describes runs on a Modal account with a self-hosted Fish Audio voice server on a GPU (about $4/hour for an H100 while it is up; it scales to zero about 5 minutes after the last request), fal (FAL_KEY, for speech-to-text, MiniMax voices and the fal MCP server), an LLM account — Cerebras by default (CEREBRAS_API_KEY, or any OpenAI-compatible endpoint I choose instead) — and, optionally, TypeSafe (TYPESAFE_API_KEY) for action choice. You also set NPCVOICE_TOKEN (the value in your own npcvoice-auth Modal secret) and NPCVOICE_URL (your voice server's URL; treat it as a secret, because its /health needs no token and wakes the GPU). The companion tooling the skill points to also needs Git, Python 3.10+, uv and ffmpeg, and the GTA V example needs Windows and the .NET SDK 8+. Do not create any account, buy anything, set any key, deploy any server or install any companion tool during this install: those are my decisions, later, if I ever ask. Install "make-npcs-talk" for me. It is a playbook for making NPCs in a game talk back: the player speaks or types, the NPC answers in character, in a cloned voice, about 1-2 seconds later, and can act on what was said. It covers the whole loop in build order (who is being talked to, scene facts, speech-to-text, an LLM in character, Fish Audio voices on a GPU, streamed playback at the NPC, subtitles, mouth movement and actions), the four hooks a game has to provide, latency budgets, costs, testing in the real game, and its rules. Repository: https://github.com/christos-antono/NPCTalk (the skill is the .agents/skills/make-npcs-talk/ folder at the pinned commit only — the identical copies under .claude/skills/ and skills/ are duplicates, not the source for this listing) Pinned version: install the code at commit 59635ceb8ee010599c3caf9d9d1ee1aa114af91d exactly. Do not substitute a newer version. If this commit is unreachable, stop and tell me. 1. Clone https://github.com/christos-antono/NPCTalk into a temporary folder, check out commit 59635ceb8ee010599c3caf9d9d1ee1aa114af91d, and copy the .agents/skills/make-npcs-talk/ folder from that checkout. Then compare every file in the folder with the fingerprints recorded for this listing on 2026-10-19: [FINGERPRINTS] - SKILL.md: SHA-256 528b406bda71ecad2a43d5d1a7f4c2c567b945bea259003ecc8d238e012d3041 - references/architecture.md: SHA-256 28cd1d93bff57b9ddea4252ffd6fd9774131a38fa78307472005591ce6cf8bda - references/game-hooks.md: SHA-256 fef64fcbea7517119dc396473c0e7ada68db2ccdc3f2cbe494a26a5de9bb5cfe - references/safety.md: SHA-256 8495c252cb080e103c9571585870c5c06063a97001909e09ca55d0bdb257655f If a file does not match, it changed since the listing was prepared: do not install it yet, tell me what you found and let me decide. 2. Read SKILL.md and its three references (references/architecture.md, references/game-hooks.md, references/safety.md) in full and tell me in plain words: what the skill does when used, which services it connects to, whether it reads or sends any personal information or keys, and anything that seems unusual. Stop on any red flag and tell me before going further. Confirm these points from the files: - The skill folder holds four markdown files only: SKILL.md (7,713 bytes) plus the three references above (3,724, 4,662 and 2,631 bytes; 18,730 bytes in total). There is no code, no script and no dependency list in the folder itself: it is instructions and reference notes for building the talking-NPC loop into a game. - Nothing is sent anywhere during installation itself; the only download at install time is the skill folder from GitHub at the pinned commit. No account, API key, telemetry or remote service is involved in copying these files. - At use time — never during this install, and only in a separate step I ask for — the loop the skill describes sends data to external services, and I want each send stated before it happens: my microphone audio (or a recording) goes to fal speech-to-text (the NVIDIA Nemotron ASR model on fal, FAL_KEY; the repo also points at the fal MCP server at mcp.fal.ai); what I said, plus the NPC's persona, a sample of the character's own lines, scene facts and per-NPC history, goes to the LLM (Cerebras api.cerebras.ai by default, or another OpenAI-compatible endpoint I choose); the NPC's reply line and a voice name go to my own Fish Audio voice server self-hosted on Modal (a *.modal.run URL in NPCVOICE_URL, with NPCVOICE_TOKEN), which streams the cloned voice back; and, if actions are enabled, what was said goes to TypeSafe Jev (api.typesafe.ai, TYPESAFE_API_KEY) to pick an action. Making a voice sends reference recordings I have the rights to to that voice server's Modal volume and transcribes them with fal. Alternatives named in the files — Fish Audio or MiniMax voices via fal (a MiniMax clone is listed at $1.50 per voice), ElevenLabs for pre-rendered lines, Whisper for speech-to-text — are used only if I pick them, never by default. Keeping the voice server warm pings it while a game runs, so the GPU stays up and keeps billing (about $45-50 for 12 hours) until it scales to zero. - Companion pieces of the repository are NOT installed by this listing: the seven other skills (fish-voice-server, voice-references, npc-dialogue, npc-actions, voice-input, voice-likeness-eval, npc-showcase-clips), the npctalk CLI, the server/ voice-server code, the GTA V example mod and the fal MCP configuration (.mcp.json). The skill points to them, and standing any of them up — deploying a Modal server, installing the CLI (which needs Python 3.10+, uv and ffmpeg), or installing files into a game folder — is a separate step, on my explicit ask, with the accounts, keys and costs above. Do not use the repository's plugin or `npx skills add` routes for this install: they install the whole repository and all of its skills, not this one folder. - The rules in the files: single-player games, or servers I run, only — never inject into online games with anti-cheat, and never bypass anti-cheat, DRM or ownership checks; clone only voices I have the rights to (my own, an actor who agreed, a licensed or generated voice) and never publish or commit voice clips, cloned renders, game audio or a game's dialogue; keys live in environment variables only, never in files; and ask me first before driving my keyboard and mouse in a running game, installing into game folders, keeping a GPU up, or publishing anything. 3. When I say "yes, install", install it as a skill: copy the whole make-npcs-talk folder into the agent's skills directory, in a folder named "make-npcs-talk", keeping all four files and their references/ structure. 4. Verify with no network calls: frontmatter valid, files in place. 5. Report what was installed, where, and what I still need to do myself (for example: create the Modal, fal, Cerebras and optional TypeSafe accounts and set FAL_KEY, CEREBRAS_API_KEY, TYPESAFE_API_KEY, NPCVOICE_TOKEN and NPCVOICE_URL in environment variables myself, stand up the companion skills, CLI and voice server myself if I want them, provide a game and voices I have the rights to myself; using the skill is a separate step that only starts when I ask for it, and the use-time sends disclosed above happen only then, to the services named). GitHub is optional: if I have a GitHub account or the gh CLI, you may use it; otherwise public access is fine. Never require it unless it's in the prerequisites above. Rules: don't touch anything outside the temp folder and the install target. If anything looks off, stop and ask me.