Tool integration · Codex

How Foundry surfaces inside OpenAI's Codex CLI.

Foundry currently supports four tools, all with the same full-skill install: Claude Code (.claude/skills/), Cursor and Codex (shared .agents/skills/), and OpenCode (.opencode/skills/). Other IDEs are not currently supported.

What the installer writes

When you select Codex during npx @engineering11/e11-foundry install, the installer writes complete skill directories into:

.agents/skills/
├── e11-master/
│   ├── SKILL.md
│   └── customize.toml
├── e11-agent-analyst/
│   ├── SKILL.md
│   └── customize.toml
├── ...              # one directory per installed skill — agents, workflows, tasks

Codex uses the shared .agents/skills/ convention (same path as Cursor). Each skill is self-contained — SKILL.md, its customize.toml, and any step files and templates it loads at run time. Older Codex installs may have files under .codex/prompts/ or ~/.codex/prompts/ — the installer cleans those up on update.

Invoking a skill

Codex reads .agents/skills/ and surfaces each directory as an invokable name:

> e11-master                                   # the entry point — catalog menu + routing
> use the e11-agent-analyst skill              # become the Analyst
> skill:e11-prd                                # run a workflow directly

Start with e11-master when you don't know what you need — it presents the installed catalog and routes. e11-help ([BH]) works from anywhere.

Picking a command

Same 2–3 letter command codes as the rest of Foundry: CB, PR, CA, DS, CR, etc. Type the code, the long name, or describe what you want — fuzzy matching is built into each agent's menu.

Full catalog: first-run.md or .foundry/_config/e11-help.csv.

Shared with Cursor

Because Codex and Cursor both read from .agents/skills/, installing Foundry for Codex makes the same skills available to Cursor (and vice versa) in the same project. The installer detects this and won't double-write. If you switch tools mid-project, no re-install is needed.

Ancestor-directory conflict check

Like the other supported tools, the installer refuses to write a Foundry install that conflicts with one in a parent directory. Pick one location and stick to it.

Headless

The e11 agent / e11 pipeline CLI drives the claude and opencode runtimes directly; Codex isn't a driven runtime. For Codex-based automation, use print mode to get the assembled prompt and pipe it into your own runner:

e11 agent dev "implement the next ready story" --runtime print

Tips

• @engineering11/alloy reads from the project's node_modules/, same as in any other tool. Install alloy with npm install @engineering11/alloy in the same project where you run Codex.
• Fresh context for [CR] Code Review. Open a new Codex session for review rather than continuing the implementation session.
• Long workflows. Phase 4 story cycles can be lengthy. Plan for the context budget; the quick flow is the lighter alternative.
• Customize from .foundry/custom/. Files under .agents/skills/ are overwritten on update; your overrides survive. See Customization.

What this page does not cover

• Authentication or billing for Codex itself.
• Codex-specific prompt customisation or context-window tuning.

Source

Platform config: tools/cli/installers/lib/ide/platform-codes.yaml (see the codex: entry).

See also: Installation · First run · Claude Code · OpenCode.