persona-voice
Skill2 starsChanged 51 days agoMITView on GitHub
Give a persona an ElevenLabs voice (one per language, auto-selected by traits) so /persona-talk replies can be spoken via [speak]. Use when the user runs /persona-voice, types [speak], or asks about persona voices in plain language.
- Reads credentials
1 other project uses a version of this.
Adopted by
Most recently taken up by
- astrowicked/devils-counciltheir own version, first seen Jun 4, 2026
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- astrowicked/devils-council★ 2same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository31 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- .github/copilot-instructions.mdCopilot instructions
- affinitySkill
- ai-personaSkill
- analytics-syncSkill
- backlogSkill
- catalogSkill
- cold-startSkill
- contradictionsSkill
- demo-dataSkill
- dovetail-syncSkill
- exportSkill
- extract-findingsSkill
- feature-panelSkill
- forget-participantSkill
- graph-lintSkill
- interview-guideSkill
- journey-mapSkill
- opportunity-treeSkill
- persona-avatarSkill
- persona-querySkill
- persona-talkSkill
- persona-workshopSkill
- prdSkill
- prioritizeSkill
- researcherSkill
- retentionSkill
- source-syncSkill
- subject-accessSkill
- welcomeSkill
The file
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 3 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 36
Remove/replace: a real participant's real name, employer, exact street/neighborhood, phone/email, or any other detail that could identify a …
- Distinctive: line 38
Only the persona's spoken words get anonymized-then-sent — never send the researcher-context block, frontmatter, or any file metadata to …
- Distinctive: line 43
Run it through the anonymization step above.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
persona-voice — Give a Persona an ElevenLabs Voice
Connects a persona to one or more ElevenLabs voices — one per language she's actually spoken in — so /persona-talk replies can be heard, not just read. Auto-selects a voice matching the current conversation's language and the persona's traits, keeps every persona vocally distinct from every other, and generates audio on demand — never automatically, never without the anonymization step below.
Requires ELEVENLABS_API_KEY in .env (already .gitignored in this repo). If it's missing, say so plainly and stop — never ask the user to paste a key into chat.
Trigger
/persona-voice Personas/Emma.md ← configure (or re-configure) a voice for this persona
[speak] ← mid-conversation in /persona-talk: voice the last replyAlso loads whenever the user asks about persona voices in plain language ("can Emma have a voice?", "change Jake's voice", "what does Tom sound like?") — this isn't limited to the two triggers above. Offered once during /welcome's Step 4 setup, but just as available any time after — configuration is never a one-time locked-in choice.
Configuration — one voice per language, auto-selected, always reported, always changeable
A persona can end up with several voices — one per language she's actually been voiced in, not just one fixed voice overall. The same persona answering in English one session and Polish the next should sound like two different, language-native voices, not one voice stretched across languages by a multilingual model — that matches ElevenLabs' own guidance that a native voice per language gives the best result. Stored as a map on the Persona file (voices:), populated lazily: only the languages actually used get an entry, nothing upfront.
- Detect which language is needed right now — the language of the current conversation (
persona-talkalready adapts to whatever language the user is writing in), not necessarily the language herTranscripts/were recorded in. A persona whose source data is all in English can still need a Polish voice if that's what this session is in. - Check her
voices:map first. An entry already there for this language means: use thatvoice_id, nothing else to do — this is what makes the choice stick once made. - Missing entry → auto-select, no confirmation prompt. Search the full shared voice library (
GET /v1/shared-voices, filterable bylanguage/locale/gender/descriptive/age) rather than just the ~28 default voices sitting on the account (GET /v1/voices) — the default set is generic/stock-sounding by comparison; the shared library'sdescriptive/use_casefields (e.g. "casual", "conversational") match a persona's actualTone of Voicefar better than the handful of preset options. A shared-library voice can be used directly in a text-to-speech call by itsvoice_id, no separate "add to account" step needed. Filter by the language plus whatever the Persona file implies about her (apparent age, gender,Tone of Voiceregister). Never pick avoice_idalready used by a different persona — read every other file inPersonas/and exclude anything already claimed anywhere in theirvoices:maps, so personas stay vocally distinct from each other. (The same persona reusing a similar-sounding voice_id across her own different languages isn't a conflict — that check is cross-persona, not cross-language.) - Always say what got picked, right after. One line is enough: "Emma now has a Polish voice too (separate from her English one) — want a different one? Just tell me a specific voice, or ask me to try again." Never silently pick and move on.
- Persist it — add or update that language's entry in her
voices:map (seePersonas/_template.md). Data like everything else in this graph, not a hidden setting. - Changing it later is a normal request, not a special flow — "give Emma a different Polish voice" replaces just that one language entry; her other languages are untouched. "Use voice
<id>for Tom's English" sets it directly, no filtering needed.
Before any text reaches ElevenLabs — anonymize, every time
This is a hard rule, not a courtesy: strip identifying details from the text before it ever leaves the repo, whether it's a live [speak] reply or a one-off demo clip.
- Remove/replace: a real participant's real name, employer, exact street/neighborhood, phone/email, or any other detail that could identify a real person — swap in the fictional persona's own name or a generic term instead.
- For this repo's current demo content, this is close to a no-op — Emma/Jake/Tom are already fictional stand-ins with no real PII in their lines. The rule exists so this stays correct the day real participant data flows through the same path; don't skip writing/checking it just because today's pass-through is trivial.
- Only the persona's spoken words get anonymized-then-sent — never send the researcher-context block, frontmatter, or any file metadata to the API. ElevenLabs only ever sees a line of in-character dialogue, nothing else.
[speak] — voice the last reply
- Take the persona's most recent in-character reply only — never the researcher-context block, never a footnote line.
- Run it through the anonymization step above.
- Look up the
voice_idfor this conversation's current language in hervoices:map. If that language has no entry yet, run the Configuration step above first (it's a normal part of[speak], not a separate blocking step the user has to remember to do beforehand). - Call ElevenLabs' text-to-speech endpoint with that
voice_id, usingeleven_v3as the default model — the most expressive/natural option on this account, clearly better thaneleven_multilingual_v2for how alive a reply sounds. It responds to inline audio tags ([laughs],[sighs], etc.) placed directly in the text — use one only where the reply already implies that beat (e.g. a stage direction like "(laughs)" in the source text becomes[laughs]), never invented for effect. Confirm the exact current endpoint/parameters against ElevenLabs' live docs regardless — API surfaces drift, don't assume a hardcoded shape is still current. - Save the returned audio to
.claude/voice-cache/(gitignored — binary audio never belongs in this git-based markdown graph) and tell the user the file path so they can play it.
Cost awareness
ElevenLabs is a metered, paid API. [speak] is opt-in per-reply by design — never suggest generating audio for a whole conversation's worth of replies at once, and never trigger it automatically just because the researcher-context block escalated to full tier.
One-off demo clip
The same [speak] pipeline, run once against a real representative English exchange, produces a standalone audio file meant for the user to upload to ElevenLabs themselves however they want it hosted. Not an automated or repeating task — just this same mechanism, used once, on request.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

