fix: ChatBubble crash and DeepSeek API compatibility

- Fix ChatBubble to handle non-string content with String() wrapper
- Fix API route to use generateText for non-streaming requests
- Add @ai-sdk/openai-compatible for non-OpenAI providers (DeepSeek, etc.)
- Use Chat Completions API instead of Responses API for compatible providers
- Update ChatBubble tests and fix component exports to kebab-case
- Remove stale PascalCase ChatBubble.tsx file
This commit is contained in:
Max
2026-01-26 16:55:05 +07:00
parent 6b113e0392
commit e9e6fadb1d
544 changed files with 113077 additions and 427 deletions

View File

@@ -0,0 +1,116 @@
---
name: 'v-02a-validate-metadata'
description: 'Validate metadata and append to report'
nextStepFile: './v-02b-validate-persona.md'
validationReport: '{bmb_creations_output_folder}/validation-report-{agent-name}.md'
agentMetadata: ../data/agent-metadata.md
agentFile: '{agent-file-path}'
---
# Validate Step 2a: Validate Metadata
## STEP GOAL
Validate the agent's metadata properties against BMAD standards as defined in agentMetadata.md. Append findings to validation report and auto-advance.
## MANDATORY EXECUTION RULES
- 📖 CRITICAL: Read the complete step file before taking any action
- 🔄 CRITICAL: Read validationReport and agentMetadata first
- 🔄 CRITICAL: Load the actual agent file to validate metadata
- 🚫 NO MENU - append findings and auto-advance
- ✅ YOU MUST ALWAYS SPEAK OUTPUT In your Agent communication style with the config `{communication_language}`
### Step-Specific Rules:
- 🎯 Validate metadata against agentMetadata.md rules
- 📊 Append findings to validation report
- 🚫 FORBIDDEN to present menu
## EXECUTION PROTOCOLS
- 🎯 Load agentMetadata.md reference
- 🎯 Load the actual agent file for validation
- 📊 Validate all metadata fields
- 💾 Append findings to validation report
- ➡️ Auto-advance to next validation step
## MANDATORY SEQUENCE
**CRITICAL:** Follow this sequence exactly. Do not skip, reorder, or improvise unless user explicitly requests a change.
### 1. Load References
Read `{agentMetadata}`, `{validationReport}`, and `{agentFile}`.
### 2. Validate Metadata
Perform these checks systematically - validate EVERY rule specified in agentMetadata.md:
1. **Required Fields Existence**
- [ ] id: Present and non-empty
- [ ] name: Present and non-empty (display name)
- [ ] title: Present and non-empty
- [ ] icon: Present (emoji or symbol)
- [ ] module: Present and valid format
- [ ] hasSidecar: Present (boolean, if applicable)
2. **Format Validation**
- [ ] id: Uses kebab-case, no spaces, unique identifier
- [ ] name: Clear display name for UI
- [ ] title: Concise functional description
- [ ] icon: Appropriate emoji or unicode symbol
- [ ] module: Either a 3-4 letter module code OR 'stand-alone'
- [ ] hasSidecar: Boolean value, matches actual agent structure
3. **Content Quality**
- [ ] id: Unique and descriptive
- [ ] name: Clear and user-friendly
- [ ] title: Accurately describes agent's function
- [ ] icon: Visually representative of agent's purpose
- [ ] module: Correctly identifies module membership
- [ ] hasSidecar: Correctly indicates if agent uses sidecar files
4. **Agent Type Consistency**
- [ ] If hasSidecar: true, sidecar folder path must be specified
- [ ] If module is a module code, agent is a module agent
- [ ] If module is 'stand-alone', agent is not part of a module
- [ ] No conflicting type indicators
### 3. Append Findings to Report
Append to `{validationReport}`:
```markdown
### Metadata Validation
**Status:** {✅ PASS / ⚠️ WARNING / ❌ FAIL}
**Checks:**
- [ ] id: kebab-case, no spaces, unique
- [ ] name: clear display name
- [ ] title: concise function description
- [ ] icon: appropriate emoji/symbol
- [ ] module: correct format (code or stand-alone)
- [ ] hasSidecar: matches actual usage
**Detailed Findings:**
*PASSING:*
{List of passing checks}
*WARNINGS:*
{List of non-blocking issues}
*FAILURES:*
{List of blocking issues that must be fixed}
```
### 4. Auto-Advance
Load and execute `{nextStepFile}` immediately.
---
**Validating persona...**