- Home
- Skills
- Agents & Orchestration
- Yuanjing Storyforge — Comic Drama Production
Works with the AI Agents you already use
Yuanjing Storyforge — Comic Drama Production
Manifest-first agent skill for AI comic-drama production: shot contracts, provider-neutral generation plans, cost gates, SRT and assembly checks, and auditable human review. Runs fully offline — no paid provider calls.
Free
Yuanjing Storyforge — Comic Drama Production
Example session with this skill installed
Plan episode 1 of a vertical comic drama from this 2-shot script, estimate the cost, and give me everything I need before I spend money on a video provider. No paid calls.
Shot 1 (5s): rain on a rooftop at night, a girl hears a signal from an old radio. Dialogue: 你也听见了吗?
Shot 2 (6s): close-up of the radio dial turning by itself, static resolving into a voice.
- Read your context and instructions
- Compiled the yuanjing storyforge
Summary
Validated a 2-shot, 11-second vertical (9:16) episode manifest and produced the full pre-generation package offline. Status: READY_FOR_PROVIDER_REVIEW. No provider was called.
What ran
python3 scripts/quickstart.py
PASS: demo-ep01 (2 shots)
Deliverables
| Artifact | Meaning |
|---|---|
previz.html | text storyboard: shot, camera, action, dialogue, end state |
episode.srt | subtitles generated from shot dialogue |
assembly.json | ffmpeg assembly plan — status: blocked, 2 clips missing |
qa_result.json | per-shot review sheet — PENDING_HUMAN_REVIEW, publish_approval: false |
run_state.json | resumable run state: task/status only, no prompts, no signed URLs |
quickstart.json | artifact index with absolute paths |
Subtitle output
1
00:00:00,000 --> 00:00:05,000
你也听见了吗?
The part that matters
The blocked assembly and the pending review are the point: the skill refuses to call content publishable just because a command succeeded. A human approves each asset by hash before delivery.
Next steps
- Fill
release_metadata(platform, AI disclosure, copyright basis) and re-runvalidate_manifest.py --strict. - Choose a provider adapter, set an approved budget, then execute explicitly.
Connects securely to your tools. The creator never sees your data.
About this skill
What it is
Yuanjing Storyforge (元境·漫界工坊) is a manifest-first production skill for AI comic dramas and short-form serialized video. It does not pretend to be an end-to-end video generator. It gives your agent the part that actually breaks in production: a validated shot contract, a provider-neutral generation plan, explicit cost gates, subtitle and assembly planning, and auditable human review.
What you get
| Capability | Command | Output |
|---|---|---|
| Validate a shot manifest | scripts/validate_manifest.py | strict schema + continuity check |
| First offline run (2 shots) | scripts/quickstart.py | storyboard HTML, SRT, assembly plan, QA report |
| Resumable mock runner | scripts/run_mock.py | run state, zero paid calls |
| Per-second cost estimate | scripts/estimate_cost.py | budget before you spend |
| Subtitle planning | scripts/make_srt.py | SRT from shot dialogue |
| Assembly plan | scripts/assemble_plan.py | ffmpeg plan, blocked until real clips exist |
| Technical QA | scripts/qa_report.py | checklist that never auto-approves |
| Publish gate | scripts/approve_publish.py | hashes every asset a human approved |
Design commitments
- Offline by default. Examples and CI never call a paid provider.
- Explicit opt-in for generation. Provider adapters are separate; an ambiguous submission is recorded as
unknownand never auto-retried, so no duplicate paid work. - No fake approvals. QA stays
PENDING_HUMAN_REVIEW, assembly staysblockeduntil real clips exist. Those are safeguards, not failures. - Provider-neutral. You keep the manifest; you choose the video model.
Requirements
Python 3.11+ and the standard library for everything above. ffmpeg/ffprobe optional, for a disposable MP4 and technical QA. Works from Claude Code, Codex CLI, Cursor, Copilot, Gemini CLI, or any agent that reads SKILL.md.
Source
Apache-2.0, v1.9.3: https://github.com/263311487-ux/yuanjing-storyforge
Seedance adapter and legacy audio are not production-validated. No auto-retries for ambiguous tasks. Reviewer names are self-reported. Real publication requires external identity controls.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 6 days ago
- Free with an Agensi account
Requires