Codex CLI¶
Codex CLI gets full learnship capabilities including real parallel subagents and the complete workflow suite.
Install¶
Installs to ~/.codex/learnship/.
Invoke commands¶
Codex CLI uses a $learnship- prefix (dollar sign + hyphen):
$learnship-ls
$learnship-new-project
$learnship-discuss-phase 1
$learnship-plan-phase 1
$learnship-execute-phase 1
$learnship-verify-work 1
$learnship-quick "fix the login bug"
$learnship-help
$learnship-review # two-pass review: spec compliance + quality (v2.4.0)
$learnship-ship # v2.0: test → commit → push → PR
$learnship-compound # v2.0: capture solved problem as knowledge
$learnship-challenge # v2.0: stress-test scope
$learnship-ideate # v2.0: codebase-grounded idea generation
Skills¶
Skills are installed as context files:
~/.codex/learnship/skills/
├── agentic-learning/
│ ├── SKILL.md
│ └── references/
└── impeccable/
├── SKILL.md
└── [21 sub-skills]/
Reference explicitly to invoke:
Parallel subagents¶
Codex CLI runs parallel subagents by default. Every new project created with $learnship-new-project starts with parallelization.enabled: true.
To disable and run sequentially:
Capabilities¶
| Feature | Status |
|---|---|
| Slash commands | ✅ $learnship-* prefix |
@agentic-learning skill | ✅ Context file |
impeccable skill suite | ✅ Context file |
| Parallel subagents | ✅ on by default |
| Wave execution | ✅ on by default |
| Agent personas (17) | ✅ Task() subagents + inline <persona_context> |
| Interactive questions | ✅ request_user_input |
| Playwright MCP smoke tests | ✅ Via playwright/mcp MCP server |
Playwright MCP smoke tests¶
Live UI smoke tests via Playwright MCP are supported when @playwright/mcp is configured. The /verify-work and /ship workflows will use it automatically for UI verification when available.
Tip
Codex CLI uses $learnship- (dollar sign prefix) rather than /learnship: or /learnship-. This matches Codex CLI's native skill invocation convention.
Tip
AGENTS.md is not auto-loaded on Codex CLI the way it is on Windsurf or Claude Code. Run /new-project once per project — it generates an AGENTS.md at your project root. Codex CLI reads AGENTS.md when commands explicitly reference it, but for fully automatic context loading, keep AGENTS.md in the project root and your agent will encounter it via the workflows.