Files

Max e9e6fadb1d 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

2026-01-26 16:55:05 +07:00

2.9 KiB

Raw Blame History

critical_actions

Activation instructions that execute every time the agent starts.

Purpose

Numbered steps that execute FIRST when an agent activates.

Use for:

Loading memory/knowledge files
Setting file access boundaries
Startup behavior (greeting enhancement, data fetch, state init)
Any MUST-do activation behavior

Applies to: BOTH Simple and Expert agents

Expert Agent Pattern

# ✅ CORRECT Expert Agent
critical_actions:
  - 'Load COMPLETE file {project-root}/_bmad/_memory/journal-keeper-sidecar/memories.md'
  - 'Load COMPLETE file {project-root}/_bmad/_memory/journal-keeper-sidecar/instructions.md'
  - 'ONLY read/write files in {project-root}/_bmad/_memory/journal-keeper-sidecar/'
  - 'Search web for biotech headlines from last 2 days, display before menu'

CRITICAL Path Format:

{project-root} = literal text (not replaced)
Sidecar created next to agent.yaml during BUILD, then copied to _memory/ during BMAD INSTALLATION
Use {project-root}/_bmad/_memory/{sidecar-folder}/ format for RUNTIME paths in agent YAML

Simple Agent Pattern

# ✅ CORRECT Simple Agent with activation behavior
critical_actions:
  - 'Give user an inspirational quote before showing menu'
  - 'Review {project-root}/finances/ for most recent data file'

Note: Agents without activation needs can omit critical_actions entirely.

Path Reference Patterns

Type	Pattern
Expert sidecar	`{project-root}/_bmad/_memory/{sidecar-folder}/file.md`
Simple data	`{project-root}/finances/data.csv`
Output folders	`{output_folder}/results/`

critical_actions vs principles

critical_actions	principles
Technical activation steps	Philosophical guidance
"Load memories.md"	"I believe in evidence"
MUST execute on startup	Guides decision-making

Grey area: "Verify data before presenting" can be either - activation behavior vs philosophical belief. Use judgment.

What the Compiler Adds (DO NOT Duplicate)

Load persona
Load configuration
Menu system initialization
Greeting/handshake

Your critical_actions become numbered steps AFTER compiler initialization.

Common Issues

Wrong Path Format

# ❌ WRONG
- 'Load ./journal-keeper-sidecar/memories.md'

# ✅ CORRECT
- 'Load COMPLETE file {project-root}/_bmad/_memory/journal-keeper-sidecar/memories.md'

Missing COMPLETE Keyword

# ❌ WRONG
- 'Load file memories.md'

# ✅ CORRECT
- 'Load COMPLETE file {project-root}/_bmad/_memory/journal-keeper-sidecar/memories.md'

COMPLETE ensures LLM reads entire file, not a portion.

Duplicating Compiler Functions

# ❌ WRONG - compiler does these
- 'Load my persona'
- 'Initialize menu system'
- 'Say hello to user'

# ✅ CORRECT - agent-specific only
- 'Load memory files'
- 'Search web for headlines before menu'

2.9 KiB Raw Blame History