onboard-agent

Original：🇺🇸 English

Translated

Use when starting a new project, adding a new agent to an existing system, or setting up workflow infrastructure from scratch.

5installs

Sourcesharpdeveye/maestro

Added on2026-04-11

NPX Install

npx skill4agent add sharpdeveye/maestro onboard-agent

SKILL.md Content

View Translation Comparison →

MANDATORY PREPARATION

Invoke {{command_prefix}}agent-workflow — it contains workflow principles, anti-patterns, and the Context Gathering Protocol. Follow the protocol before proceeding — if no workflow context exists yet, you MUST run {{command_prefix}}teach-maestro first.

Bootstrap a new agent workflow from scratch, or add a new agent to an existing system.

Step 1: Establish Conventions

markdown

## Workflow Conventions
### Prompt Format
- Delimiter style: [XML tags / markdown headers / triple-dash]
- Section order: [System → Context → Instructions → Input]
- Output format: [JSON with schema / markdown template]

### Tool Conventions
- Naming: [verb_noun / noun.verb / camelCase]
- Description template: [What → When → When Not → Returns]
- Error format: [{ code, message, details }]

### Logging
- Format: [JSON structured]
- Required fields: [workflow_id, step, timestamp, level]

### File Structure
- Prompts: [prompts/workflow-name/v1.md]
- Tools: [tools/tool-name.{ext}]
- Config: [config/environment.yaml]
- Tests: [tests/workflow-name/]

Step 2: Create Initial Structure

text

project/
├── prompts/          # System prompts, versioned
├── tools/            # Tool definitions
├── config/           # Environment-specific configuration
├── tests/            # Golden test sets and evaluation suites
├── logs/             # Runtime logs (gitignored)
└── .maestro.md       # Workflow context

Step 3: Create the First Agent

System prompt: Role definition with constraints
2-3 essential tools: Start with the minimum viable tool set
Output schema: Define expected output format
One golden test: At least one test case with known-good output
Basic error handling: Structured error responses
Logging: Structured log output for each run

Step 4: Verify

Run the agent with the golden test case
Verify error handling works (send bad input)
Verify logging captures useful context

Recommended Next Step

After onboarding, run

{{command_prefix}}diagnose

for a baseline health check, then

{{command_prefix}}fortify

to add production-grade error handling.

NEVER:

Start building without establishing conventions
Create tools without descriptions
Skip the golden test case
Over-scope the initial agent (start minimal, amplify later)