Zerops Platform Agent System

Guides AI agents operating within the Zerops deployment environment.

Core Principle

Dev is for iterating and fixing. Stage is for final validation.

Agents must test and fix errors on dev before deploying to stage. HTTP 200 doesn't mean the feature works—check response content, logs, and browser console.

Architecture

CLAUDE.md (67 lines)          → Platform fundamentals, always in context
    │
    ▼
workflow.sh --help            → Full reference, contextual phase guidance
verify.sh / status.sh         → Self-documenting tools

Knowledge lives in tools via --help, surviving context compaction and subagent spawning.

Files

zcp/
├── CLAUDE.md              # Entry point for agents
└── .claude/
    ├── workflow.sh        # Phase orchestration + help system
    ├── verify.sh          # Endpoint testing with evidence
    ├── status.sh          # Deployment status + wait mode
    └── settings.json      # Bash permissions

Workflow

INIT → DISCOVER → DEVELOP → DEPLOY → VERIFY → DONE

DEVELOP phase enforces the iteration loop:

1. Build & run on dev
2. Test functionality (not just HTTP status)
3. Check for errors (logs, responses, browser console)
4. If errors → Fix → Go to step 1
5. Only when working → deploy to stage

Usage

# Start workflow
/var/www/.claude/workflow.sh init      # Enforced (gates)
/var/www/.claude/workflow.sh --quick   # No enforcement

# Get help
/var/www/.claude/workflow.sh --help
/var/www/.claude/workflow.sh --help develop

Key Features

• Evidence-based gates - JSON files with session tracking block invalid transitions
• Self-documenting - All tools have --help with contextual guidance
• Context-resilient - Tools always available regardless of context state
• Platform gotchas - Troubleshooting table, pre-deployment checklist, inline warnings