agentleFS
hongyuanchongyuanc/codex-game-studios/.agents/skills/onboard/SKILL.md

onboard

Skill5 starsChanged 4 months agoMITView on GitHub

Use when a contributor or agent needs a role-specific summary of project state, architecture, conventions, and priorities.

We split this skill into the rules it gives an agent

10 rules. 0 are common, 5 are distinctive, and 5 are very specific to this project.

  • 0

    Common

    Most of the 148 edited versions we read keep these.

  • 5

    Distinctive

    This author's own choices. Worth a look.

  • 5

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Before default delegation, run python3 ../../../tools/codex_studio/agent_delegation.py resolve --project-root <project-root> --role <role> …

    line 9
  2. Distinctive:

    Ask one decision question per turn and wait for the answer before asking another.

    line 13
  3. Very specific:

    Use request_user_input for structured choices when it is available.

    line 14
  4. Distinctive:

    Use Codex custom agents by role and profile when delegation is useful.

    line 15
  5. Distinctive:

    Treat any approved write as one complete proposed changeset.

    line 16
  6. Distinctive:

    Read AGENTS.md for project overview and standards.

    line 25
  7. Very specific:

    Read the relevant agent definition from .codex/agents/ if a specific role is specified.

    line 27
  8. Distinctive:

    Read recent changes (git log if available) to understand current momentum.

    line 39
  9. Very specific:

    Run $sprint-status to show the new contributor current progress.

    line 103
  10. Very specific:

    Run $help if the contributor needs guidance on what to work on next.

    line 104

“Common” is measured: kept by at least 6 in 10 of the 148 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md
<!-- codex-studio-delegation: governed -->

Resolve every role through ../../../.codex/docs/plugin-agent-delegation.md; do not require a repository-local .codex/agents/ or .codex/agent-packs/ tree. Before default delegation, run python3 ../../../tools/codex_studio/agent_delegation.py resolve --project-root <project-root> --role <role> and use only its returned role contract.

Codex Interaction Contract

  • Ask one decision question per turn and wait for the answer before asking another.
  • Use request_user_input for structured choices when it is available; otherwise ask the same concise question directly.
  • Use Codex custom agents by role and profile when delegation is useful.
  • Treat any approved write as one complete proposed changeset. Before writing, present every target path and material edit together; do not add unlisted files or behavior.
  • A new path, expanded scope, or material change requires a revised complete proposed changeset and fresh approval.

Output Safety

Generate the onboarding draft in conversation. Before writing, show the complete proposed changeset and target path and ask once for approval.

Phase 1: Load Project Context

Read AGENTS.md for project overview and standards.

Read the relevant agent definition from .codex/agents/ if a specific role is specified.


Phase 2: Scan Relevant Area

  • For programmers: scan src/ for architecture, patterns, key files
  • For designers: scan design/ for existing design documents
  • For narrative: scan design/narrative/ for world-building and story docs
  • For QA: scan tests/ for existing test coverage
  • For production: scan production/ for current sprint and milestone

Read recent changes (git log if available) to understand current momentum.


Phase 3: Generate Onboarding Document

# Onboarding: [Role/Area]

## Project Summary
[2-3 sentence summary of what this game is and its current state]

## Your Role
[What this role does on this project, key responsibilities, who you report to]

## Project Architecture
[Relevant architectural overview for this role]

### Key Directories
| Directory | Contents | Your Interaction |
|-----------|----------|-----------------|

### Key Files
| File | Purpose | Read Priority |
|------|---------|--------------|

## Current Standards and Conventions
[Summary of conventions relevant to this role from AGENTS.md and agent definition]

## Current State of Your Area
[What has been built, what is in progress, what is planned next]

## Current Sprint Context
[What the team is working on now and what is expected of this role]

## Key Dependencies
[What other roles/systems this role interacts with most]

## Common Pitfalls
[Things that trip up new contributors in this area]

## First Tasks
[Suggested first tasks to get oriented and productive]

1. [Read these documents first]
2. [Review this code/content]
3. [Start with this small task]

## Questions to Ask
[Questions the new contributor should ask to get fully oriented]

Phase 4: Complete Proposed Changeset

Show the onboarding draft and canonical path production/onboarding/onboard-[role]-[date].md. Ask once whether to approve that complete proposed changeset before writing. Write only that approved file.

Phase 5: Next Steps

Verdict: COMPLETE — onboarding document generated.

  • Share the onboarding doc with the new contributor before their first session.
  • Run $sprint-status to show the new contributor current progress.
  • Run $help if the contributor needs guidance on what to work on next.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

Reviews can't be read right now.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.