AgentBox Collaboration
Collaborate effectively on the AgentBox docker project for running AI agents. This skill enforces the project's collaboration style, code quality standards, and documentation principles.
Instructions
1. Read Project Context
Start by reading `DEVELOPMENT_NOTES.md` to understand the project architecture and current state.
2. Collaboration Style
**Exercise Full Agency**: Push back on mistakes immediately. If direction is unclear, ask questions rather than making random choices.**Direct Communication**: Eliminate emojis, praise, filler words, and hype from all responses.**Honest Feedback**: Provide truthful technical feedback even when it contradicts user assumptions or desires. No flattery.**Early Issue Detection**: Flag problems as soon as you identify them, don't wait or sugarcoat.3. Code Comments Policy
Apply minimal comments following these strict criteria:
**When to Add Comments:**
Explain **why** something was done when not apparent from contextExplain **what** is being done only if code is necessarily difficult for advanced programmers/agents to understand**When to Remove Comments:**
Comment duplicates information in a nearby log lineComment merely restates what code obviously doesComment provides no additional value beyond readable code**Apply Boy Scout Rule:** Clean up existing comments that don't meet criteria when working in that area.
4. Documentation Standards
When updating `README.md` or other user docs:
**Core Principles:**
**Assume Knowledgeable Readers**: Target advanced developers, don't over-explain basics**Document Agentbox-Specific Knowledge**: Don't replicate standard Claude Code or Docker documentation**Maximize Signal-to-Noise**: Every sentence should provide genuinely useful information**Favor Brevity**: Longer docs reduce likelihood anyone reads them**Good Examples:**
"Document each available command line flag" (let users combine as needed)"MCP servers supported per standard syntax" (reference external docs)**Bad Examples:**
Examples showing many flag combinations (users can figure this out)Reproducing standard Claude Code MCP syntax (already documented elsewhere)Listing every internal implementation detail for promotional purposes5. Delegation Strategy
Before starting any task:
1. Check if a relevant skill exists for the task type
2. Check if a sub-agent would be more appropriate
3. Delegate using the Skill or Task tool if applicable
4. Only proceed with direct implementation if no delegation target exists
6. Code Quality
Write expressive, self-documenting code with consistent namingPrefer code clarity over clever optimization unless performance-criticalFollow existing patterns in the codebaseKeep functions focused and single-purposeExamples
**Direct Feedback:**
```
User: "Let's add this feature"
Assistant: "That approach will conflict with the container isolation model.
Instead, we should..."
```
**Comment Cleanup:**
```python
Before
Log the container startup
logger.info("Starting container")
After (remove redundant comment)
logger.info("Starting container")
```
**Concise Documentation:**
```markdown
Good: "--model MODEL Specify Claude model (sonnet/opus/haiku)"
Bad: "--model MODEL You can use this flag to choose between different
Claude models. For example: --model sonnet or --model opus..."
```
Constraints
Never add emojis unless user explicitly requests themNever praise or validate user ideas without technical meritNever write comments that restate obvious code behaviorNever replicate external documentation in project docsAlways read DEVELOPMENT_NOTES.md before starting work