peon-ping-create-pack

You are drafting ONE brand-new PeonPing pack from scratch. Work only inside the draft directory you create. Do not touch any other directory, do not install anything, do not commit anything, and never finish by declaring the pack "done" or "installed" — a freshly created pack is never installed directly (see Hard rules).

If invoked without name/flavor/vibe/draft_root

This skill is internal: it is only ever invoked headlessly by peon create with name, flavor, vibe, and draft_root already filled in. If you were invoked WITHOUT all four (a human ran this directly with missing arguments), do nothing except tell them to run peon create instead. Do not touch any files. Exit.

Procedure

1. Parse the invocation: name (already validated by the caller against

^[a-z0-9][a-z0-9_-]*$), flavor (sfx or voice), vibe (a one-line description), draft_root.

1. Compute draft_dir = <draft_root>/<name>. If it already exists, STOP —

tell the human a draft with that name already exists and to run peon eval <name> on it (never overwrite an existing draft). Otherwise create draft_dir/sounds/.

1. Author a concept honoring the vibe, covering all 7 CESP categories

with at least one sound each: session.start, task.acknowledge, task.complete, task.error, input.required, resource.limit, user.spam. This mirrors the concept-authoring discipline of the brand-to-peon-packs skill's step 2:

• Every sound needs a label — short, caption-quality text describing

what's heard (not a filename, not a restatement of the category — e.g. "Soft acknowledgement", "Serene completion", not "task_complete_0").

• Keep the aesthetic consistent across all 7 sounds so the pack reads as

one coherent idea, not seven unrelated clips. Let the vibe string drive instrument palette / tone / register (e.g. "calm bells" implies soft chimes and bowls throughout, never harsh or percussive).

• flavor: sfx — wordless. Each sound needs a prompt: a concrete,

renderable description for ElevenLabs sound-generation (instrument or texture, character, brief shape — e.g. "two gentle bells, soft decay, warm room tone"). Avoid vague adjectives with nothing to render ("nice sound") — describe what would actually produce the sound.

• flavor: voice — spoken. Each sound needs text (an in-character

line fitting the category and vibe) plus a voice_id (an ElevenLabs voice id). Reuse ONE voice_id across all 7 sounds — a pack has one voice, not seven. If no voice was specified in the invocation, pick a stock ElevenLabs voice whose character plausibly fits the vibe; do not invent a voice_id string — use one you can confirm actually exists (list voices via the ElevenLabs API if unsure).

1. Write draft_dir/openpeon.json — a CESP manifest with cesp_version,

name, display_name, version (start at "0.0.1"), a short description reflecting the vibe, and "x_openpeon_draft": true (always — a pack authored by this skill is never anything but a draft). categories maps each of the 7 category names to {"sounds": [{"file": "sounds/<category>_<index>.wav", "label": "..."}]}.

1. Write draft_dir/prompts.json — the contract the peon-ping-remix skill

reads for rerolls: {"<file>": {"type": "sfx"|"tts", "prompt": "..."}} for sfx sounds, or {"<file>": {"type": "tts", "text": "...", "voice_id": "..."}} for voice sounds. Keys are the same file paths used in openpeon.json (relative to draft_dir, e.g. sounds/session_start_0.wav).

1. Render every sound. For each entry, write a small job file under

draft_dir/jobs/ (e.g. draft_dir/jobs/render-job-<category>_<index>.json — never at the draft root; approve prunes everything under jobs/, so a render input left at the root would ship as junk in the approved pack) shaped as {"type": "sfx"|"tts", "prompt"|"text"+"voice_id", "out": "<absolute path to the WAV>"}, then run: python3 <peon-ping>/scripts/pack-render.py --job <file> Pass --mock when the environment variable PEON_RENDER_MOCK=1 is set — this writes a silent placeholder WAV instead of calling ElevenLabs, so the whole draft can be authored and rendered with no network, no API key, and no ffmpeg. This is how automated tests exercise this skill. (Resolve <peon-ping> as the scripts directory next to the running peon install, in the same order the peon-ping-remix skill uses: $PEON_DIR/scripts when set, else ${CLAUDE_CONFIG_DIR:-$HOME/.claude}/hooks/peon-ping/scripts, else the repo checkout you were invoked from.)

1. The renderer exits nonzero on a malformed job, a missing key, or a

silent-after-retry render. If ANY sound fails to render, STOP — print the renderer's stderr and exit nonzero. Do not leave openpeon.json claiming sounds exist that never rendered; it is fine to leave the partial draft_dir on disk for the human to inspect or retry — do not delete it.

1. After every sound renders successfully, print a short summary (pack name,

flavor, category count, sound count) and always end by telling the human to run: peon eval <name> to listen to and approve the draft. Never claim the pack is finished, ready to use, or installed — creation only ever produces a draft that still has to clear the eval gate.

Hard rules

• NEVER write outside draft_dir.
• NEVER set x_openpeon_draft to anything but true.
• NEVER install the pack, copy it into a packs directory, or otherwise make

it directly usable — that only happens via peon eval <name> → approve.

• NEVER print or log an ElevenLabs key.
• Honor PEON_RENDER_MOCK=1 by passing --mock to every pack-render.py

call — this is load-bearing for tests and for previewing the flow without spending API credits.

• The flow ALWAYS ends by telling the human to run peon eval <name> —

creation never ends with an installed pack.