ai-maestro-programmer-agent

AI Maestro Programmer Agent (AMPA)

A general-purpose, multi-language programmer agent for Claude Code that implements, tests, fixes, and documents code across Python, JavaScript/TypeScript, Rust, Go, and other toolchains — either standalone or orchestrated within the AI Maestro ecosystem. The current version is recorded in .claude-plugin/plugin.json (the single source of truth).

Overview

The AI Maestro Programmer Agent is a general-purpose programmer that executes implementation tasks assigned by the Orchestrator. It handles the actual coding work across multiple programming languages and toolchains.

Standalone or Orchestrated: This plugin works in two modes. In standalone mode, the agent receives tasks directly from the user and reports back in conversation. In orchestrated mode (within the AI Maestro ecosystem), it receives tasks from the Orchestrator (AMOA) via messaging. No additional setup is needed for standalone mode.

Implementer Agents

In the AI Maestro ecosystem, implementers are agents that produce artifacts. The Programmer Agent (AMPA) is one subtype of implementer — it produces code, tests, and pull requests. Other implementer subtypes include artists (visual assets), SFX experts (audio assets), and more. All implementers share the same role (implementer) in team registries but use subtype-specific plugins and naming (e.g., svgbbox-programmer-001).

Prefix: ampa- = AI Maestro Programmer Agent

Core Responsibilities

1. Code Implementation: Write and modify source code according to specifications
2. Test Writing: Create comprehensive test suites
3. Code Fixing: Resolve bugs and linting/type errors
4. Documentation: Write inline documentation and docstrings
5. Multi-Language Support: Work across Python, JavaScript, Rust, Go, and compiled languages

Components

Agent (1)

Agent	File	Description
ai-maestro-programmer-agent-main-agent	agents/ai-maestro-programmer-agent-main-agent.md	Main general-purpose programmer agent

Skills (6)

Skill	Description
ampa-task-execution	Execute programming tasks per requirements
ampa-orchestrator-communication	Communication with the Orchestrator (AMOA) agent
ampa-github-operations	Git and GitHub operations (clone, branch, commit, PR)
ampa-project-setup	Initialize project configuration and install tooling
ampa-handoff-management	Create and receive handoff documents and bug reports
ampa-prrd-trdd-kanban	The MEMBER (programmer) role in the PRRD / TRDD / kanban workflow

Memory (global, janitor-hosted)

This plugin uses the global AI-Maestro markdown memory system — it ships no per-plugin memory skills or rules/ mirror. Recall / write / update go through the global janitor-memory-recall / -write / -update skills; the protocol + recall law live in ~/.claude/rules/markdown-memory-recall.md; the project's memory contract + scope routing live in CLAUDE.md. The git-tracked PROJECT-scope wiki is .claude/project/memory/ (stood up once via /janitor-memory-bootstrap). The memgrep binary (from ai-maestro-janitor) powers recall and degrades to grep when absent — recall degrades, never breaks.

Hooks

None. The hooks/hooks.json is empty -- AMPA uses globally installed hooks.

Scripts

The scripts/ directory contains 4 project utility scripts. Plugin validation runs through the CPV remote launcher (uvx … cpv-remote-validate), fetched on demand from Emasoft/claude-plugins-validation — the previously vendored CPV validator scripts were retired (CI and publish.py both call the remote validator, so local copies only drifted behind upstream).

Script	Description
publish.py	Strict release pipeline — test, lint, validate, bump, push
pre-push-hook.py	Git pre-push hook — runs cpv-remote-validate before each push
test_order_pipeline.py	Tests publish.py release-step ordering
smart_exec.py	Cross-platform script executor with timeout support

Token-Efficient Reporting

Project scripts support file-based reporting to minimize terminal output: