AlexandrAI AGENTS.md
agents-md

Gumroad Orchestrator Role Agent Guide

Project-specific AGENTS.md guidance for the Gumroad orchestrator role, manual role sequence, execution ledger, dry-run and publish modes, and post-publish shorts handoff.

AGENTS.md

Guidance for coding agents working on the Gumroad orchestrator role.

Project Overview

The orchestrator coordinates one manual Gumroad AI execution kit run. It applies Markdown role files as specialist lenses, writes product files, runs deterministic scripts, and stops. It must not create background services, wrapper daemons, databases, cron jobs, or autonomous bulk publishing loops.

Source Files

  • agents/orchestrator.md defines the run controller.
  • skills/product-orchestrator/SKILL.md defines the manual product run.
  • AGENTS.md defines global workspace boundaries and reporting shape.
  • roadmap/product-roadmap.md defines current product lanes and brand position.
  • skills/taste-driven-product-judgment/SKILL.md is the global judgment gate.
  • scripts/product_artifacts.py, scripts/gumroad_cli.py, and scripts/shorts_artifacts.py provide deterministic execution.

Setup Commands

bashpython3 scripts/gumroad_status.py snapshot --output-dir research
UV_LINK_MODE=copy UV_PROJECT_ENVIRONMENT="$HOME/.cache/gumroad-codex-agent-venv" uv run python scripts/product_artifacts.py render products/<slug>/product.json
UV_LINK_MODE=copy UV_PROJECT_ENVIRONMENT="$HOME/.cache/gumroad-codex-agent-venv" uv run python scripts/product_artifacts.py qa products/<slug>
python3 scripts/gumroad_cli.py dry-run products/<slug>

Use publish and shorts commands only when the owner explicitly requested publish mode and QA passed.

Change Rules For Agents

  • Read AGENTS.md, agents/orchestrator.md, and roadmap/product-roadmap.md before product planning.
  • Run the read-only status snapshot when Gumroad CLI login is available.
  • Apply the role order from orchestrator: status, CEO, market, planner, content, support, page, QA, publisher.
  • In publish mode, continue into video strategist, scriptwriter, Hyperframe, TTS, shorts QA, and social metadata only after Gumroad publish succeeds.
  • Record Agent Invocation Ledger with execution mode, role file, skill file, inputs, output section, and evidence.
  • Say markdown-role-applied unless a live subagent was actually spawned.
  • Include Taste Gate with buyer decision moment, kept proof path, cut distraction, and trust reason.
  • Ask only when missing detail affects safety, account mutation, or publishability.

Verification

bashUV_LINK_MODE=copy UV_PROJECT_ENVIRONMENT="$HOME/.cache/gumroad-codex-agent-venv" uv run python scripts/product_artifacts.py render products/<slug>/product.json
UV_LINK_MODE=copy UV_PROJECT_ENVIRONMENT="$HOME/.cache/gumroad-codex-agent-venv" uv run python scripts/product_artifacts.py qa products/<slug>

QA checks meeting.md for Agent Invocation Ledger, required role paths, global taste skill, execution mode, and Taste Gate. Fix the meeting trace before treating QA as complete.

Security Notes

Do not publish account credentials, buyer data, private endpoints, local absolute paths, CLI auth state, or hidden operational notes. Keep public docs repo-relative and workflow-focused.

PR Checklist

  • Run mode is clear: draft, dry-run, or publish.
  • Product run has meeting.md and product.json.
  • Agent Invocation Ledger is honest.
  • Taste Gate is specific.
  • Publish happens only after explicit owner intent and QA.