jogg-lip-sync（慢跑唇形同步）

Jogg Lip Sync

Use this skill to execute lip sync tasks directly, not to generate integration code.

All paths in this document are relative to the current skill root directory.

Runner:

• sh "run.sh"

Trigger

• User asks to run lip sync
• User asks to check lip sync task status
• User provides video and audio and expects the final driven video result

Required Inputs

• video input: URL or local file path
• audio input: URL or local file path

Optional:

• JOGG_BASE_URL, default https://api.jogg.ai
• JOGG_API_PLATFORM, default openclaw
• playback_type, default normal
• poll_interval_seconds, default 10
• max_wait_seconds, default 1800

If any required input is missing, ask only for the missing item.

Default values used when unset:

• JOGG_BASE_URL=https://api.jogg.ai
• JOGG_API_PLATFORM=openclaw
• JOGG_LIP_SYNC_DEFAULT_PLAYBACK_TYPE=normal
• JOGG_LIP_SYNC_DEFAULT_POLL_INTERVAL_SECONDS=10
• JOGG_LIP_SYNC_DEFAULT_MAX_WAIT_SECONDS=1800

• JOGG_API_KEY is required.
• Other current environment variables are optional.
• If JOGG_API_KEY is empty, stop and tell the user to purchase an API plan at https://www.jogg.ai/api-pricing/ and obtain an API key before continuing.

Hard Rules

• Execute the existing runner in the current run.
• Prefer the fixed runner over handwritten HTTP calls.
• Use run.sh as the only runner entrypoint.
• Do not write scripts, helper files, SDKs, wrappers, or temporary executors.
• Do not replace execution with code generation.
• Do not create duplicate tasks for the same normalized inputs in one run.
• Reuse existing tasks whenever allowed by the decision rules.
• Prefer returning the final video result over producing artifacts.

Endpoints

• POST /v2/upload/asset
• GET /v2/lip_sync_video
• POST /v2/create_lip_sync_video
• GET /v2/lip_sync_video/:task_id

Header:

• X-Api-Key: $JOGG_API_KEY
• optional x-api-platform: $JOGG_API_PLATFORM

Procedure

1. Collect missing inputs only.
2. For create or reuse flow, execute the runner with --no-poll first.
3. Parse the returned JSON and read task_id plus status.
4. If status is pending or processing, call the runner again with --task-id to query or poll.
5. Return the execution result directly in the conversation.

Output contract:

• stdout: final machine-readable JSON result only
• stderr: progress logs during upload, query, create, and polling
• Recommended agent pattern: create with --no-poll, then query by task_id

Runner Modes

Create or reuse a task:

sh "run.sh" \
  --video "<video-url-or-file>" \
  --audio "<audio-url-or-file>" \
  --playback-type "normal" \
  --no-poll

Query a task by task_id:

sh "run.sh" \
  --task-id "<task-id>"

Useful flags:

• --force-recreate: only when the user explicitly asks to regenerate after a terminal task
• --poll: wait until terminal state in task_id mode
• --no-poll: return immediately in create or reuse mode; recommended for the first runner call from the skill
• --poll-interval-seconds
• --max-wait-seconds

run.sh behavior:

• Uses the native shell implementation directly.
• Requires curl and jq.
• Uses the system default values when optional environment variables are unset.

Decision Rules

• playback_type defaults to normal if omitted.
• The query endpoint returns the latest matching task under the current authenticated user and space.
• Reuse pending, processing, and success tasks by default.
• Do not recreate a failed task unless the user explicitly requests a retry.
• Query before every create attempt.

Allowed playback_type values:

• normal
• normal_reverse
• normal_reverse_by_audio

Output

Return only execution results:

• action
• reused
• Whether an existing task was reused or a new one was created
• task_id
• Current status
• data.result_url when successful
• error.message when failed
• If still running, return the live task_id and status
• Never replace the result with a generated script or file