@mmmbuto/nexuscrew 0.9.0 → 0.9.2

package/skills/alibaba-token-media/SKILL.md

@@ -1,152 +0,0 @@
----
-name: alibaba-token-media
-description: "Generate or edit images with Wan 2.7 and create videos with HappyHorse through an Alibaba Qwen Cloud Token Plan Personal subscription. Use when the user asks Claude Code, Codex, Codex-VL, or Pi to draw, generate, or edit an image with wan2.7-image or wan2.7-image-pro; create text-to-video or image-to-video media with happyhorse-1.1-t2v or happyhorse-1.1-i2v; inspect a submitted HappyHorse task; or validate the Token Plan media setup. Do not use this skill for chat, visual understanding, batch automation, backend services, or pay-as-you-go DashScope APIs."
----
-
-# Alibaba Token Media
-
-Use the bundled, dependency-free CLI to access only the media models included
-with the Lite guardrails of Alibaba Token Plan Personal. The same skill and Python script work
-with Claude Code, Codex, Codex-VL, and Pi. Invoke the script directly from Pi;
-do not assume that Pi provides a native MCP client.
-
-Keep this workflow interactive. Every generation consumes plan Credits, and
-the plan is not for unattended batch or backend workloads.
-
-## Safety contract
-
-- Read the credential only from `ALIBABA_CODE_API_KEY`. Never request it in
-  chat, pass it on the command line, print it, persist it, or send it through
-  an agent bus.
-- Use only the fixed Token Plan host and exact model allowlist implemented by
-  the script. Never fall back to DashScope pay-as-you-go endpoints.
-- Run `status` and then a `--dry-run` before the first real request in a
-  session. Dry runs never require or reveal the key.
-- Show the user the normalized model, operation, size or duration, resolution,
-  and confirmation flags before spending Credits. A real image request needs
-  `--confirm-credit-use`; a real video request additionally needs
-  `--confirm-expensive`.
-- Require `--confirm-high-cost` for Wan Pro, 2K or 4K images, 1080P video, or
-  video longer than five seconds. Keep Lite defaults at one 1K base image or
-  one 720P three-second video.
-- Never batch, enable sequential image sets, retry a generation POST
-  automatically, or run more than one generation submit concurrently. A task
-  status GET may be retried, but never with a tight polling loop.
-- Accept local input only as a regular, non-symlink JPEG, PNG, BMP, or WEBP
-  file below 20 MiB and under the current user's home. Image-to-video accepts
-  JPEG, PNG, or WEBP first frames, not BMP. Accept remote input only
-  over public HTTPS; reject localhost, private IP literals, credentials, and
-  non-standard ports.
-- Download signed result URLs immediately. Never paste signed URLs into chat,
-  logs, source files, or an agent bus; they expire after 24 hours.
-
-## Locate the CLI
-
-Resolve `scripts/alibaba_token_media.py` relative to this `SKILL.md`, then run
-it with Python 3:
-
-```bash
-python3 <skill-dir>/scripts/alibaba_token_media.py status
-```
-
-Keep this packaged skill as the source of truth; do not copy the script into a
-project before using it.
-
-## Generate or edit an image
-
-Start with a sanitized preview:
-
-```bash
-python3 <skill-dir>/scripts/alibaba_token_media.py image \
-  --prompt "A quiet alpine lake at sunrise" \
-  --dry-run
-```
-
-After explicit approval, repeat without `--dry-run` and add
-`--confirm-credit-use`. For an edit, add one to nine `--image` values. Each
-value may be a permitted local image under the current user's home or a public
-HTTPS URL:
-
-```bash
-python3 <skill-dir>/scripts/alibaba_token_media.py image \
-  --prompt "Preserve the subject; replace the background with a night city" \
-  --image /absolute/path/input.png \
-  --confirm-credit-use
-```
-
-Use `--model wan2.7-image-pro --size 2K --confirm-high-cost` only after the
-user selects that higher-cost variant. `4K` is allowed only for Pro
-text-to-image with no input images. The script fixes `n=1` and disables image
-sets. Image watermarking is off by default and enabled only with `--watermark`.
-
-## Submit a video
-
-Text-to-video Lite preview:
-
-```bash
-python3 <skill-dir>/scripts/alibaba_token_media.py video-submit \
-  --prompt "A paper model city slowly lights up at dusk" \
-  --dry-run
-```
-
-For a real request, add both `--confirm-credit-use` and
-`--confirm-expensive`. For image-to-video, select `happyhorse-1.1-i2v` and
-supply exactly one first-frame image with `--image`. Do not supply `--ratio`
-for image-to-video; its aspect ratio follows the input. Video watermarking is
-on by default and can be disabled explicitly with `--no-watermark`.
-
-```bash
-python3 <skill-dir>/scripts/alibaba_token_media.py video-submit \
-  --model happyhorse-1.1-i2v \
-  --image /absolute/path/first-frame.png \
-  --prompt "The camera slowly moves forward" \
-  --confirm-credit-use --confirm-expensive
-```
-
-Return the task ID after a real submit. Do not hide submission success behind
-polling.
-
-## Check and download a video
-
-Query a task once:
-
-```bash
-python3 <skill-dir>/scripts/alibaba_token_media.py video-status <task-id>
-```
-
-When it is `SUCCEEDED`, repeat with `--download`. If recurring monitoring is
-requested and the client exposes a loop facility, use an interval of at least
-five minutes. Do not implement a private tight polling loop.
-
-```bash
-python3 <skill-dir>/scripts/alibaba_token_media.py video-status <task-id> \
-  --download
-```
-
-Default downloads go to
-`~/Downloads/alibaba-token-plan/YYYY-MM-DD/` with unique names and no
-overwrite. An explicit `--output` must remain under the current user's home.
-
-## Reference
-
-Read [references/api-contract.md](references/api-contract.md) before changing
-payloads, limits, endpoints, or model handling.
-
-## Dependencies
-
-**Bundled (installed with the package):** the CLI
-`scripts/alibaba_token_media.py` ships with this skill and is
-dependency-free — Python 3 standard library only (verified against the
-script's imports; it needs a Unix-like environment because it uses `fcntl`,
-so Windows is out of scope).
-
-**External (you must provide):**
-
-| Need | Install | Probe / failure mode |
-|---|---|---|
-| Python 3 | Debian/Ubuntu `apt install python3`, Fedora `dnf install python3`, macOS `brew install python`, Termux `pkg install python` | `python3 --version` fails → nothing in this skill runs |
-| An Alibaba Cloud **Token Plan Personal** subscription that includes the media models | activate on the Alibaba Cloud console — no local install | API calls fail with an authorization error from the service |
-| API key in the environment | `export ALIBABA_CODE_API_KEY=<your key>` in the shell/service that runs the script | the script refuses with `ALIBABA_CODE_API_KEY is not set in this process environment` — the failure names the exact variable; no network call is made without it |
-
-The key is read only from the environment: never paste it into a prompt, a
-script argument, or a file.

package/skills/alibaba-token-media/agents/openai.yaml

@@ -1,4 +0,0 @@
-interface:
-  display_name: "Alibaba Token Media"
-  short_description: "Generate Wan images and HappyHorse videos"
-  default_prompt: "Use $alibaba-token-media to generate or edit media safely with my Alibaba Token Plan."

package/skills/alibaba-token-media/references/api-contract.md

@@ -1,97 +0,0 @@
-# Alibaba Token Plan media API contract
-
-Verified against the official Qwen Cloud documentation on 2026-07-19.
-
-## Fixed subscription surface
-
-- Host: `https://token-plan.ap-southeast-1.maas.aliyuncs.com`
-- Credential environment variable: `ALIBABA_CODE_API_KEY`
-- Authorization: `Bearer <token>`
-- Supported invocation surfaces: Claude Code, Codex, Codex-VL, and Pi through
-  the packaged skill and CLI. Pi is not assumed to support MCP natively.
-- Token Plan Personal Lite guardrails: interactive use only, one submit at a
-  time, no batch generation, and no automatic retry of a generation POST.
-- Never substitute the DashScope pay-as-you-go host.
-
-The Token Plan documentation says coding tools cannot configure media models
-as chat models. They must use a skill, command, agent, or equivalent extension.
-The Personal plan is intended for interactive programming and agent-tool
-work, not unattended scripts, backend services, or batch processing.
-
-## Wan 2.7 image generation and editing
-
-Synchronous endpoint:
-
-`POST /api/v1/services/aigc/multimodal-generation/generation`
-
-Allowed plan models:
-
-- `wan2.7-image`
-- `wan2.7-image-pro`
-
-Request shape:
-
-```json
-{
-  "model": "wan2.7-image",
-  "input": {
-    "messages": [{
-      "role": "user",
-      "content": [{"text": "prompt"}, {"image": "data-or-https-url"}]
-    }]
-  },
-  "parameters": {
-    "n": 1,
-    "size": "1K",
-    "enable_sequential": false,
-    "watermark": false,
-    "thinking_mode": true
-  }
-}
-```
-
-The content array requires exactly one text object and accepts zero to nine
-image objects. Prompt maximum is 5,000 characters. Inputs may be JPEG, PNG,
-BMP, or WEBP and are limited to 20 MiB each. `wan2.7-image` supports 1K and 2K;
-Pro also supports 4K for text-to-image only. The output image URL is under
-`output.choices[*].message.content[*].image` and expires after 24 hours.
-
-## HappyHorse video generation
-
-Submit endpoint:
-
-`POST /api/v1/services/aigc/video-generation/video-synthesis`
-
-Required submission header: `X-DashScope-Async: enable`.
-
-Allowed plan models:
-
-- `happyhorse-1.1-t2v`
-- `happyhorse-1.1-i2v`
-
-Text-to-video input contains `prompt`; parameters contain `resolution`,
-`ratio`, and `duration`. Image-to-video input additionally contains exactly one
-`media` item with `type: first_frame` and an HTTPS or Base64 image in `url`;
-its parameters omit `ratio` because output follows the first frame. The CLI
-enables video watermarking by default; image watermarking is disabled by
-default. Image-to-video first frames accept JPEG, PNG, or WEBP, not BMP.
-
-Duration is an integer from 3 through 15 seconds. Resolution is 720P or 1080P.
-Text-to-video ratios are 16:9, 9:16, 1:1, 4:3, 3:4, 4:5, 5:4, 9:21, or 21:9.
-The response task ID is `output.task_id`.
-
-Status endpoint:
-
-`GET /api/v1/tasks/{task_id}`
-
-Terminal states are `SUCCEEDED`, `FAILED`, `CANCELED`, and `UNKNOWN`. On
-success, `output.video_url` is an H.264 MP4 signed URL valid for 24 hours.
-
-## Official sources
-
-- https://docs.qwencloud.com/token-plan/best-practices/integrate-multimodal-gen
-- https://docs.qwencloud.com/token-plan/personal/token-plan-personal-overview
-- https://docs.qwencloud.com/api-reference/image-generation/wan27-image-gen-edit/synchronous
-- https://docs.qwencloud.com/api-reference/video-generation/happyhorse-text-to-video/create-task
-- https://docs.qwencloud.com/api-reference/video-generation/happyhorse-image-to-video/create-task
-- https://docs.qwencloud.com/api-reference/video-generation/happyhorse-text-to-video/query-result