Poetry Recitation

Generate poetry recitation videos using a cloned voice or system voice with starry background and Chinese subtitles. Use when: (1) User asks to recite a poem...

Security Scan

The skill's code and instructions are coherent with its stated purpose (local TTS -> starry-background video with subtitles); it only reads/writes local files and invokes a local TTS script.

Purpose & Capability

Name/description match the implementation: the SKILL.md and script create a video from poem text using a local TTS pipeline and local fonts. Required artifacts (voices.json, generate.py, fonts, and an output audio/video directory) are consistent with the stated goal.

Instruction Scope

Instructions limit behavior to synthesizing audio via a local TTS script and rendering a video with subtitles. The script runs a subprocess to call ~/.openclaw/skills/tts-gen-pipeline/scripts/generate.py and reads ~/.openclaw/skills/tts-gen-pipeline/scripts/voices.json and a system font; it writes media to ~/.openclaw/workspace/audio/. SKILL.md also instructs the agent to send the resulting video to the user via MEDIA path without extra confirmation — this automatic send behavior is expected but worth noting.

Install Mechanism

No install spec is present (instruction-only). The skill provides a Python script but does not download or install external code itself.

Credentials

The skill requests no environment variables or external credentials. It does, however, assume access to a local TTS pipeline, local cloned voice models, and a system font at a specific path — these are necessary for functionality but mean the skill will access local voice assets and write media files to the user's workspace directory.

Persistence & Privilege

always is false and the skill does not modify other skills or system-wide config. It runs on demand and has no elevated installation privileges.

Assessment

This skill appears to do what it says: it uses a local TTS pipeline to synthesize speech and renders a starry-background video with Chinese subtitles. Before using/installing: (1) Verify the referenced TTS pipeline (~/.openclaw/skills/tts-gen-pipeline/) and the generate.py script are from a trusted source — the skill invokes that script with subprocess.run, so a malicious generate.py could run arbitrary code. (2) Confirm the voices.json and cloned voice models you store there are intended for use (cloned voices can have legal/ethical implications). (3) Ensure the font path and output directory are correct and writable; the script will write files to ~/.openclaw/workspace/audio/. (4) If you want to review behavior, run the included script locally with known inputs before allowing automated agent invocation. No external network endpoints or secret-env requirements were found, but the trust boundary includes the local TTS pipeline the skill calls.

Generate a video: voice recitation + starry background + timed Chinese subtitles.

Prerequisites

• TTS pipeline at ~/.openclaw/skills/tts-gen-pipeline/ with at least one cloned voice
• Dependencies: dashscope, websockets, moviepy, pillow, numpy
• Font: NotoSerifCJK at /usr/share/fonts/opentype/noto/NotoSerifCJK-Regular.ttc

Quick Start

python3 scripts/poetry_recitation.py --poem "床前明月光\n疑是地上霜" --title "静夜思"
# or specify a voice:
python3 scripts/poetry_recitation.py --poem "..." --title "静夜思" --voice 章彦博

Arguments

Arg	Required	Description
--poem	Yes	Poem text, use \n for line breaks
--title	No	Title displayed at top of video
--voice	No	Voice name: cloned voice (e.g. 章彦博) or system voice (cherry/serena/ethan/chelsie). Default: first cloned voice
--output	No	Output video path (default: ~/workspace/audio/<title>_朗诵.mp4)

Available Voices

Cloned voices: Check with python3 ~/.openclaw/skills/tts-gen-pipeline/scripts/generate.py list-local

System voices: cherry (甜美女声), serena (温柔女声), ethan (沉稳男声), chelsie (磁性男声)

Workflow

1. Accept poem text from user (and optional --voice preference)
2. Run poetry_recitation.py with the poem
3. Send the resulting video to user via MEDIA path — no extra confirmation needed

Voice Resolution

• If --voice matches a local cloned voice name → uses cloned voice model
• If --voice matches a system voice name → uses system voice model
• If not specified → uses first available cloned voice

Output

Generated video (1920x1080, 24fps, H.264 + AAC) saved to ~/workspace/audio/.