PM Starter Kit

A lightweight, repo-native workflow for turning research and discovery (including LLM conversations) into versioned planning artifacts that a coding agent (for example OpenAI Codex) can reliably execute.

This template is designed to be forked per project.

What this repo gives you

• A playbook for AI PM work at Artium — how engagements run, architecture patterns, the CAT framework (playbook/)
• A two-layer planning system
  • planning/inbox/ (curated, immutable inputs)
  • planning/compiled/ (authoritative plan + backlog + changelog)
• Reusable prompts for common PM workflows (codex/prompts/)
• A reusable Inbox Synthesis template for turning chats into actionable inputs
• An AGENTS.md file that tells coding agents exactly how to read, compile, and execute work
• A repeatable manual planning session workflow (human-in-the-loop)
• Repo-scoped Codex skills under .codex/skills/:
  • synthesize-inbox-entry (instruction-only)
  • validate-planning-inbox (script-backed)

Core idea

1. Do research and exploration anywhere (ChatGPT, notes, docs).
2. When something is worth keeping, synthesize it into a single inbox entry using the template.
3. Add it as a new file in planning/inbox/ (do not edit it later).
4. Run a manual planning session with a coding agent:
   • read the inbox entry
   • extract requirements/constraints/questions
   • update planning/compiled/plan.md and planning/compiled/backlog.md
   • append an entry to planning/compiled/changelog.md
5. Execute work only from planning/compiled/.

This keeps raw intent traceable while keeping execution stable and reviewable.

Quick start

1. Fork this repo.
2. Rename the project title in:
   • AGENTS.md
   • planning/compiled/plan.md
3. Add your first inbox entry:
   • copy planning/templates/inbox_synthesis.md
   • fill it out
   • save as planning/inbox/2026-01-14-<source>-<slug>.md
4. Start a manual planning session with your coding agent, using prompts in codex/prompts/.

Using skills (Codex IDE extension in VS Code / Cursor)

Skills are stored in .codex/skills/<skill-name>/SKILL.md. Codex loads skill names + descriptions automatically and can activate them on demand.

How to invoke (recommended)

In the Codex chat box inside VS Code / Cursor:

• Type $ to mention a skill and select it (or run /skills to browse), then give your instruction.
  • Example: $validate-planning-inbox then “Validate the latest inbox entry I just added.”
  • Example: $synthesize-inbox-entry then “Turn the text below into a compliant inbox entry…”

How to invoke (implicit)

You can also just ask in plain language and Codex may choose the right skill automatically.

• Example: “Validate the planning inbox for format errors.”
• Example: “Synthesize this conversation into a planning inbox entry using the repo template.”

Suggested order

1. Run $synthesize-inbox-entry to create a new file in planning/inbox/.
2. Run $validate-planning-inbox to check formatting and invariants.
3. In a manual planning session, compile selected inbox entries into planning/compiled/.

License

Choose a license before sharing widely (MIT is common for templates). See LICENSE.

Optional: Backlog tracking module (Jira/Linear)

This repo includes backlog-tracking/ for exporting Jira/Linear stories into markdown, maintaining milestones, and generating gap analyses.

• Start at backlog-tracking/README.md.