name: wing-models description: "Use this skill whenever the user wants to route part of a message to an OpenAI-compatible model while keeping the rest as local agent instructions. Trigger on requests like '和模型聊聊', '帮我问问模型', '代问模型'." argument-hint: "user message, alias/model profile"

Wing-Models

This skill packages a repeatable OpenAI-compatible model conversation workflow.

Alias Credential Set (Mandatory)

• Do not use a single API_KEY + MODEL_ID pair.
• Use required 4-step interactive profile input: alias -> baseurl -> apikey -> modelid.
• For every credential entry (including the first one), collect alias, baseurl, apikey, and modelid one-by-one in chat.
• baseurl is mandatory and must be provided explicitly (no default value).
• Never auto-fill first-entry alias/modelid from template defaults.
• note is optional and may be left empty; accept skip / 跳过 / - as empty note when chat UI cannot send blank messages.
• At least one profile entry must exist.
• Store profiles in .3rd.env as WING_MODELS_PROFILE_SET.
• Legacy alias:key:model text format is not supported.
• --alias is mandatory; script will error if alias is not provided.

If no profile set exists and the script is interactive, prompt user to enter profile entries.

When an agent needs to create .3rd.env directly after collecting fields in chat, use this structure:

WING_MODELS_PROFILE_SET=[{"alias":"<alias-from-chat>","baseURL":"https://openrouter.ai/api/v1","apiKey":"<api-key-from-chat>","modelId":"<model-id-from-chat>","note":""}]

Supported Providers

This skill works with any OpenAI-compatible API, including:

• OpenRouter (https://openrouter.ai/api/v1)
• OpenAI (https://api.openai.com/v1)
• Azure OpenAI (https://YOUR_RESOURCE.openai.azure.com/openai/deployments/YOUR_DEPLOYMENT)
• Local models (e.g., http://localhost:11434/v1 for Ollama)
• Any other OpenAI-compatible endpoint

Output Contract

• Save outputs directly under <cwd>/.
• Dialogue outputs: *-dialogue.md (question + answer; attachment sections record paths only)
• Attachment outputs: *-attachment-<n>.<ext> files
• Credentials file: <skill-dir>/.3rd.env
• Always print model reply immediately in chat.

Security Rules (Mandatory)

• Never print API keys in chat or terminal logs.
• Prefer interactive/env-based profile setup over command-line key arguments.
• If credentials are missing, collect in chat and persist to .3rd.env with restrictive permissions.

Large File Authorization (Mandatory)

Before reading saved model output files (.md or attachments):

• If file size is greater than 50KB (51200 bytes), ask user authorization in chat first.
• If user refuses/skips, do not read content; only report path and size.

Required Assets

• Script: ./scripts/wing_models.mjs
• Package: ./scripts/package.json
• Dependency: openai
• Dependency: dotenv
• Credential template: ./.3rd.env.template

Runtime Readiness Check (Automatic)

Before executing the script, always verify runtime prerequisites automatically:

Pre-flight Checklist

1. Check Node availability: Run node --version to verify Node.js is installed and accessible.
2. Check npm availability: Run npm --version to verify npm is installed and accessible.
3. Check dependency installation: Run npm list --prefix <skill-dir>/scripts openai dotenv to verify both dependencies are installed.

Auto-install Logic

If any check fails:

1. If node or npm is unavailable, report the missing tool and halt execution. Ask user to install Node.js first.
2. If dependencies are missing (exit code non-zero), auto-run:

   npm install --prefix <skill-dir>/scripts
   

3. Re-run the dependency check after installation.
4. If installation fails, report the error and halt execution.

Implementation Pattern

# Step 1: Check node
node --version || { echo "Node.js not found. Please install Node.js."; exit 1; }

# Step 2: Check npm
npm --version || { echo "npm not found. Please install Node.js with npm."; exit 1; }

# Step 3: Check dependencies
npm list --prefix <skill-dir>/scripts openai dotenv 2>/dev/null || npm install --prefix <skill-dir>/scripts

# Step 4: Proceed with script
node <skill-dir>/scripts/wing_models.mjs --alias <alias> --prompt "<prompt>"

One-time Install Verification

For first-time use or after clearing node_modules, the above pattern ensures dependencies are installed transparently without manual intervention.

Run Template

Install once:

npm install --prefix <skill-dir>/scripts

Call template:

node <skill-dir>/scripts/wing_models.mjs \
  --alias <alias> \
  --prompt "<user-prompt>"

Long prompt template:

node <skill-dir>/scripts/wing_models.mjs \
  --alias <alias> \
  --prompt-file <path-to-prompt.txt>

With attachment input (repeatable):

node <skill-dir>/scripts/wing_models.mjs \
  --alias <alias> \
  --prompt "<user-prompt>" \
  --attachment <path-or-url>

Reliability Notes

• For long prompts, prefer --prompt-file over shell heredoc/complex quoting.
• Treat terminal-rendered body as preview only; use [TEXT_FILE] path as source of truth for full output.
• If output seems cut off, inspect the saved file first, then request continuation in a follow-up call.
• Check [ROUTE] marker for provider/alias/model/baseURL decisions when debugging.

Completion Checks

• Runtime readiness verified: Node.js and npm are available, dependencies are installed.
• Alias selected correctly (arg or first available).
• Model reply printed immediately.
• No API key exposed in logs.
• Large-file consent requested when needed.