Auto Model Switcher

Automatically selects the best model based on task type and requirements. Use when: (1) Task requires specific capabilities (coding, analysis, multimodal, wr...

Security Scan

The skill's instructions, requirements, and scope are internally consistent for an instruction-only model-routing helper, but there are a few small operational ambiguities (logging destination and how model availability is checked) you should verify before enabling it.

Purpose & Capability

Name and description match the instructions: the SKILL.md contains model lists, selection logic by task type, fallback rules, manual overrides, and notes to validate against openclaw.json. No unrelated env vars, binaries, or installs are requested.

Instruction Scope

Instructions remain within the stated purpose (detect task type, pick a configured model, preserve session). Two ambiguous items: (1) 'Log model selections for learning and optimization' does not specify where logs are stored or sent; (2) 'Always check if target model is available' implies reading openclaw.json or querying provider status — that may require the agent to access config or provider endpoints that are outside the skill text but are reasonable operational requirements.

Install Mechanism

Instruction-only skill with no install spec and no code files — minimal install risk because nothing is fetched or written by the skill itself.

Credentials

No environment variables, credentials, or config paths are required by the skill. The SKILL.md explicitly restricts switching to pre-configured models in openclaw.json, which is proportionate. Note: actual model availability checks or provider interactions will rely on existing agent configuration and credentials outside this skill.

Persistence & Privilege

The skill is not always-enabled and does not request elevated persistence. There is no instruction to modify other skills or system-wide agent settings. Autonomous invocation is allowed by platform default but is not combined with other high-risk flags.

Assessment

This skill appears to do what it says: automatically pick a model from your pre-configured list. Before installing: (1) confirm your openclaw.json lists only provider models you trust, (2) verify where model-selection logs are written or sent (local vs remote) and whether that meets your privacy needs, (3) ensure the agent already has the provider credentials/configuration needed to check model availability, and (4) test the skill in a limited environment to confirm it preserves session context and respects manual overrides.

Intelligently selects the optimal model from available providers based on task characteristics.

When to Use

• Task requires specific capabilities (coding, analysis, multimodal, writing, research)
• Need optimal performance/cost balance
• Working with long context or complex reasoning
• User doesn't specify a model preference

Available Models Analysis

Qwen Series (bailian provider)

Model	Context	Multimodal	Best For	Cost
qwen3.5-plus	1M	✅ Text+Image	General tasks, creative writing, balanced performance	Low
qwen3-max	262K	❌ Text only	Complex reasoning, deep analysis, research	High
qwen3-coder-plus	1M	❌ Text only	Code generation, debugging	Medium

Third-party Models (bailian provider)

Model	Context	Multimodal	Best For
glm-5	1M	✅ Text+Image	Multimodal tasks, Chinese optimization
kimi-k2.5	200K	✅ Text+Image	Multimodal, research-oriented
MiniMax-M2.5	1M	✅ Text+Image	Long context multimodal

Selection Logic

Task Type Detection

Code Tasks → bailian/qwen3-coder-plus

• Keywords: code, programming, debug, fix, implement, develop, coding, script
• File extensions: .py, .js, .ts, .java, .cpp, etc.
• Commands: git, npm, docker, build, compile

Complex Analysis → bailian/qwen3-max

• Keywords: analyze, research, compare, evaluate, strategy, deep dive, business analysis
• Tasks requiring multi-step reasoning
• Financial/strategic analysis

Research Tasks → bailian/qwen3-max

• Keywords: research, investigate, study, survey, academic, literature review
• Complex information synthesis
• Multi-source analysis and comparison

Writing/Copywriting Tasks → bailian/qwen3.5-plus

• Keywords: write, draft, copywriting, content, article, blog, email, proposal, creative
• Marketing copy, social media content
• Creative writing and storytelling

Multimodal Tasks → bailian/glm-5

• Image analysis, OCR, visual understanding
• Audio processing (when supported)
• Mixed text+image inputs

Long Context → bailian/qwen3.5-plus

• Document processing > 200K tokens
• Summarization of large documents
• Historical context analysis

General Tasks → bailian/qwen3.5-plus (default)

• Chat, simple queries, basic tasks
• When no specific requirements detected

Fallback Strategy

1. Primary model selection based on task type
2. If primary model fails, fallback to qwen3.5-plus
3. If still failing, use current session model

Usage Examples

Automatic Selection

User: Help me debug this Python code
→ Model: bailian/qwen3-coder-plus

User: Analyze our Q4 financial performance vs competitors  
→ Model: bailian/qwen3-max

User: Research the latest AI trends in marketing
→ Model: bailian/qwen3-max

User: Write a compelling product description for our new service
→ Model: bailian/qwen3.5-plus

User: What's in this image?
→ Model: bailian/glm-5

User: Summarize this 500-page document
→ Model: bailian/qwen3.5-plus

Manual Override

Users can always specify models directly:

• /model bailian/qwen3-max
• Use coder model for this task

Implementation Notes

• Always check if target model is available before switching
• Preserve current session context when switching
• Log model selections for learning and optimization
• Respect user's explicit model preferences

Security Considerations

• Only switch between pre-configured models in openclaw.json
• Never attempt to use unconfigured or unknown models
• Validate model names against available list before switching

Performance Metrics

Track these metrics for continuous improvement:

• Task completion success rate by model
• Response time by model and task type
• User satisfaction feedback
• Cost per task type

This skill enables intelligent model routing without user intervention while maintaining full control when needed.

Iteration Support

• Skills can be updated via clawhub sync --all
• Version updates maintain backward compatibility
• New task types can be added without breaking existing functionality