AGENT SKILL

audio-stinger

Short (2-4s) sonic-logo / stinger generation — a notification chime, transition whoosh, or outro sting. Sibling to `audio-jingle`, routed through the same media dispatcher and provider ladder, but scoped to a single one-shot gesture instead of a full musical arc. Use when the brief wants one short sound, not a 30-90s bed.

License 许可
Apache-2.0
最近更新
2026-07-09
Stars
★ 0
Forks
0

01 前言 Front Matter

od
{ "mode": "audio", "preview": { "type": "html", "entry": "example.html" }, "surface": "audio", "scenario": "marketing", "design_system": { "requires": false }, "example_prompt": "A 2.5-second brand sonic logo for a fintech app — a single warm\nmarimba-like hit that rises and resolves, no vocals, clean tail with\nno reverb wash. Notification-safe (works at low volume on a phone\nspeaker)." }
name
audio-stinger
triggers
[ "stinger", "sonic logo", "sound logo", "notification sound", "whoosh", "outro sting", "transition sound", "提示音", "音效" ]
description
Short (2-4s) sonic-logo / stinger generation — a notification chime, transition whoosh, or outro sting. Sibling to `audio-jingle`, routed through the same media dispatcher and provider ladder, but scoped to a single one-shot gesture instead of a full musical arc. Use when the brief wants one short sound, not a 30-90s bed.

02 资源文件 2

03 SKILL.md 正文

Audio Stinger Skill

One short sound, one gesture. audio-jingle covers full musical beds, voiceover, and open-ended SFX; this skill is scoped to the 2-4 second one-shot case — a sonic logo, a notification chime, a transition whoosh, an outro sting — where the entire brief is "make one sound that lands."

Resource map

audio-stinger/
├── SKILL.md
├── example.html            ← what the resulting card looks like in Examples
└── references/
    └── checklist.md         ← P0/P1 self-review before dispatching

Workflow

Step 0 — Read the project metadata

audioKind (usually sfx or music for a stinger — speech doesn't apply to this format), audioModel, and audioDuration. Hard-cap audioDuration at 4 seconds. If the brief or the project metadata asks for something longer, say so and suggest audio-jingle instead rather than silently stretching a stinger past its format.

Step 1 — Plan the single gesture

A stinger is one idea, not a mood arc:

  • Texture: one clear sound source (a hit, a chime, a swell, a whoosh) — not a layered mix of three.
  • Envelope: attack shape (sharp vs. soft) and decay/tail length — the tail must resolve fully within the 2-4s window, no fade-to-silence cut short.
  • Register: does it need to read at low volume on a phone speaker (notification-safe) or is it for a video's louder mix (a transition whoosh, an outro hit)? State which.

State the one-sentence plan before dispatching.

Step 2 — Compose the prompt

Prefer specific texture words ("warm marimba-like hit," "glassy chime," "soft low-frequency swell") over mood words ("make it cool"). Bind audioDuration to the dispatcher parameter directly — never write "2 seconds" only in prose.

Step 3 — Dispatch via the media contract

Use the unified dispatcher — do not call provider APIs by hand:

"$OD_NODE_BIN" "$OD_BIN" media generate \
  --project "$OD_PROJECT_ID" \
  --surface audio \
  --audio-kind "<sfx|music>" \
  --model "<audioModel from metadata>" \
  --duration <audioDuration seconds, 2-4> \
  --output "<short-slug>-stinger.mp3" \
  --prompt "<assembled prompt from Step 1>"

The command prints one line of JSON: {"file": {"name": "...", ...}}. The bytes land in the project; the FileViewer renders transport controls automatically.

Step 4 — Self-check

Run references/checklist.md before calling this done.

Step 5 — Hand off

Reply with: the one-gesture plan, the filename returned by the dispatcher, and one concrete variation idea (e.g. "swap the attack from sharp to soft" rather than "make it different").

Hard rules

  • 2-4 seconds, hard cap. Longer requests route to audio-jingle.
  • One gesture, not a mood arc. No intro/chorus/outro structure — that belongs to a jingle, not a stinger.
  • Tail must resolve inside the runtime. No abrupt cut-off, no hanging reverb past the clip's own end.
  • No vocals unless the brief explicitly asks for a spoken sonic tag (e.g. a one-word brand call-out) — a stinger is usually instrumental or pure SFX.
  • State the intended playback context (notification-safe/low-volume vs. mixed into louder video audio) so the loudness/texture choice makes sense.

04 同仓库更多 Skills