ASTra takes security seriously. This document covers authorization, safe practices, and configuration.

By default, no users can execute commands. The allowed_users list must be configured.

# In discord_gateway.py
def is_user_authorized(self, user_id: str) -> bool:
    allowed = self._config.allowed_users
    if not allowed:
        return False  # Deny all if list is empty
    return user_id in allowed

{
  "orchestration": {
    "allowed_users": ["123456789012345678", "987654321098765432"]
  }
}

An authorized user can add others:

/auth add @username
/auth remove @username
/auth list

1. Enable Developer Mode in Discord Settings
2. Right-click your username → Copy ID

Every Discord command checks authorization:

@app_commands.command()
async def feature(self, interaction):
    if not self.is_user_authorized(str(interaction.user.id)):
        await interaction.response.send_message("❌ Not authorized")
        return
    # ... handle command

Control which shell commands can be executed:

{
  "orchestration": {
    "security": {
      "command_allowlist": ["git", "npm", "python", "pytest", "php", "composer"],
      "require_permission_for_shell": true
    }
  }
}

Only safe branches are indexed to prevent exposure of sensitive feature branches:

{
  "ingestion": {
    "safe_branches": ["main", "master"]
  }
}

If you checkout a non-safe branch:

⚠️ Clone successful, but indexing skipped.
Branch `feature/secret` is not in safe list ["main", "master"].
Switch to a safe branch to build the Knowledge Graph.

Sensitive files are excluded by default:

{
  "ingestion": {
    "ignore_patterns": [".env", "*.key", "*.pem", "secrets/"]
  }
}

Store securely:

export GITHUB_TOKEN="ghp_xxxx"

Never commit tokens to the repository.

export DISCORD_TOKEN="your-bot-token"

By default, ASTra uses local Ollama models:

{
  "llm": {
    "model": "ollama/qwen2.5-coder:7b",
    "host": "http://localhost:11434"
  }
}

Cloud fallback is disabled by default:

{
  "orchestration": {
    "fallback_to_cloud": false
  }
}

If enabled, ensure you trust the cloud provider with your code context.

Custom tools execute shell commands. Be aware of:

• Command injection: Unsanitized parameters
• Privilege escalation: Running as current user
• Data exposure: Commands may access sensitive files

1. Validate parameters: Use specific types
2. Limit scope: Make tools specific, not general
3. Audit logs: Commands are logged
4. Review YAML files: Treat as code

Example of a safe tool:

name: run_tests
description: Run test suite
command: pytest tests/ -v
parameters: {}  # No user input

Example of a risky tool:

name: run_command
description: Run any command
command: {cmd}  # DANGER: Arbitrary execution
parameters:
  cmd:
    type: string

All commands and tool executions are logged:

INFO: User 123456789 executed /checkout https://github.com/owner/repo
INFO: Executing custom tool 'deploy_staging' with args: {'branch': 'main'}

Configure log path:

{
  "orchestration": {
    "log_path": "./logs/astra.log"
  }
}

1. Use local LLMs for sensitive codebases
2. Restrict allowed_users to trusted developers
3. Review custom tools before deployment
4. Enable safe_branches for production
5. Rotate tokens regularly
6. Monitor logs for unusual activity