Narrator AI Skill
Let your AI agent produce movie/short-drama narration videos from a single sentence: search footage, pick templates, BGM and voices, generate scripts, and compose the final video.
SKILL.md mandates explicit user confirmation for every resource and the irreversible magic-video submission, discloses data flow (sole endpoint openapi.jieshuo.cn), credential storage and scope, and provides task verify/task budget for cost control — good safety design. Deducted for: pip dependency installed from a GitHub zip URL without hash pinning, total dependence on a closed-source commercial API whose key is obtained via private email contact, and irreversible paid operations with confirmation but no rollback mechanism.
Internally highly self-consistent: clearly distinguishes task_id vs task_order_num, path-asymmetric anchors, language-chain rules, 18 documented error codes, and a rule that failures must be surfaced to the user rather than auto-switching paths. Deducted for: static review cannot execute anything; install spec pins CLI v1.0.0 while README says v1.0.0+ and recommends git+main (version drift); references/ detail files are absent from the evidence so key paths cannot be cross-checked.
Trigger description is precise (film commentary, drama narration, dubbing, in Chinese and English), and China-mainland friendly: openapi.jieshuo.cn and the Feishu resource preview are reachable, ~100 built-in movie materials, Chinese UI. Deducted for: non-fit boundaries (e.g., non-narration video editing) only implicit, API key gated behind private email contact limiting accessibility, and no explicit anti-false-trigger conditions.
Good architecture: SKILL.md + references/ progressive disclosure, bilingual docs, MIT license, consistent version 1.0.5 across frontmatter and plugin.. Deducted for: no CHANGELOG/version history, maintenance responsibility only via a personal email, LICENSE copyright holder (jieshuo-ai) inconsistent with repo owner (NarratorAI-Studio), and missing references/ content prevents verifying the layering.
Claims full search→script→compose pipeline with budget estimation and error handling; if accurate, clear marginal value over manual CLI use. Deducted for: no execution evidence in static review, '✅ Verified' platform claims are author-only assertions, output usability is unverifiable, and the API key requires offline contact to obtain.
Provides auditable specifics (error codes, field formats, point pricing) and a platform matrix. Deducted for: all verification claims are author self-attestation with no third-party execution evidence or CI test suite covering the skill's key paths; references/ files absent, so conclusions are not independently reproducible.
- Core functionality depends entirely on a closed-source commercial API (openapi.jieshuo.cn); the API key must be obtained via private contact, and service availability/pricing are outside user control.
- magic-video submission is an irreversible paid operation (30 pts/minute); ensure the agent strictly follows the pre-submission confirmation rules.
- The pip dependency is installed from a GitHub zip URL without hash pinning — a supply-chain integrity risk; consider a hash-pinned fixed version after review.
- The references/ detail files are not included in this evidence set; verify the parameter tables and error-code tables against the live API after installation.
- LICENSE copyright holder (jieshuo-ai) differs from the repo organization (NarratorAI-Studio), and the publisher is unverified — confirm identity independently.
What does this skill do, and when should you use it?
This is an agent skill package (SKILL.md plus a references/ directory) that wraps the narrator-ai-cli command-line tool and connects it to the Narrator AI video-narration API. Once installed, the user simply says something like "create a narration video for this movie," and the agent handles searching footage, choosing a template, picking BGM and a dubbing voice, generating the script, and composing the video. The skill ships with roughly 100 ready-to-use movie materials (video + SRT), 146 BGM tracks, 63 dubbing voices, and 90+ narration style templates, so uploads are rarely needed. The README lists verified setups on OpenClaw, Windsurf, Claude Code, Cursor, and Tencent's WorkBuddy/QClaw, among others. Using it requires an application-issued API key (NARRATOR_APP_KEY), and tasks are billed in points.
The skill contains no executable code itself; it is a detailed instruction document teaching the agent how to invoke narrator-ai-cli subcommands: material list (built-in material search), task search-movie (external movie lookup), fast-writing / generate-writing (script generation), fast-clip-data / clip-data (clip data), video-composing (final render), the optional magic-video (visual template pass), plus standalone voice-clone and tts tasks. SKILL.md defines two workflow paths (Fast/Standard), a resource-confirmation protocol, a task polling pattern, and error-prevention rules (correct task_order_num handling, language-chain consistency, no autonomous path switching). The end product is a downloadable MP4 narration video.
- Film commentary creators who want to batch-produce narration shorts for popular movies without preparing footage or subtitles themselves
- Short-drama operators who need quick multi-language dubbed promo/narration videos for new releases
- Indie creators working inside Claude Code / Cursor who want one-prompt generation of a finished narrated video
- Creators with their own video + SRT who want a templated script-and-dubbing pipeline to speed up production
- Content teams needing standalone voice cloning or text-to-speech
What are this skill's strengths and limitations?
- End-to-end pipeline from footage search to finished MP4, with ~100 preloaded movie materials (video + SRT included)
- Very rigorously written SKILL.md: explicit anti-pattern rules, language-chain checks, step-by-step confirmation protocol, and handling guidance for all 18 API error codes
- Built-in 90+ narration templates, 146 BGM tracks, 63 dubbing voices, plus standalone voice cloning and TTS
- Cost estimation via task budget/task verify, with mandatory confirmation on expensive irreversible steps like magic-video (30 pts/minute)
- Verified setup instructions for multiple mainstream agent platforms in the README
- Closed-source paid API: requires an application-issued NARRATOR_APP_KEY and point-based billing dependent on vendor pricing
- Task polling required; search-movie can take 60+ seconds (Gradio backend)
- Path-asymmetric parameter passing (task_id vs order_num) between the two workflows causes 10001-class errors if mishandled — a cognitive burden for agents
- No test suite or independent quality evidence of generated output is mentioned in the source; README's "verified" claims cannot be independently checked
- Availability and data privacy of the openapi.jieshuo.cn endpoint depend on the vendor, with no SLA stated
How do you install this skill?
1) Install the CLI: pip install "narrator-ai-cli @ git+https://github.com/NarratorAI-Studio/narrator-ai-cli.git"
2) Configure the key: narrator-ai-cli config set app_key <your_app_key> (request a key via [email protected])
3) Install the skill (SKILL.md plus the entire references/ folder): for Claude Code, mkdir -p /path/to/project/.skills && git clone https://github.com/NarratorAI-Studio/narrator-ai-cli-skill.git /path/to/project/.skills/narrator-ai-cli; for OpenClaw use ~/.openclaw/skills/; for Cursor use .cursor/rules/. Requires Python 3.10+.
How do you use this skill?
After installation, trigger it with natural language, e.g. "Create a narration video for The Shawshank Redemption," "Show me what movies are available," or "Use a comedy template and generate a narration." The agent follows SKILL.md's flow, confirming material, path, BGM, voice, and template with you before each step. Note: every resource pick and any magic-video submission require your approval; use narrator-ai-cli user balance, task budget, and task verify to check funds and materials up front. Full parameter tables and error codes live in the references/ directory.