# Codex

> >

- **Type:** Skill
- **Install:** `agentstack add skill-lagz0ne-c3-skill-codex`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [lagz0ne](https://agentstack.voostack.com/s/lagz0ne)
- **Installs:** 0
- **Category:** [Agent Skills](https://agentstack.voostack.com/c/agent-skills)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [lagz0ne](https://github.com/lagz0ne)
- **Source:** https://github.com/lagz0ne/c3-skill/tree/main/.claude/skills/codex

## Install

```sh
agentstack add skill-lagz0ne-c3-skill-codex
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# Codex — cross-model delegation

`codex` (OpenAI Codex CLI, model `gpt-5.5`, `approval_policy = never` so it never blocks
on prompts) is a second coding agent that reads this repo and reasons over it precisely
(verified). Consult it for a different model's perspective — it catches what a single
model misses.

## When to consult Codex
- A well-scoped implementation chunk you want a second model to write, or to cross-check
  your own implementation against.
- An independent design/diff review — different model, different failure modes caught.
- Adversarial verification of a claim ("is X actually true in the shipped code?").

## How — always non-interactive `codex exec`
| Goal | Command |
|------|---------|
| Analyze / review (read-only, safe — the default) | `codex exec --sandbox read-only "" 2>/dev/null` |
| Let Codex EDIT files | `codex exec --sandbox workspace-write "" 2>/dev/null` |
| Independent review of the working tree | `codex review` |
| Apply Codex's last proposed diff | `codex apply` |
| Cheaper/deeper | add `-c model_reasoning_effort=low\|medium\|high\|xhigh` |
| Capture output | append `-o out.txt`, or `--json` for structured events |
| Continue a session | `codex exec resume --last ""` |

## Discipline (non-negotiable)
- **Scope tightly + hand it concrete material** — point Codex at exact `file:line`, the
  locked contract (`tasks/v11-locked-contract.md`), or the spec. It works best on
  concrete input, not vague asks.
- **`2>/dev/null`** to suppress thinking tokens unless you're debugging the run.
- **Verify every workspace-write run with the repo's own gates** — `go -C cli build ./...`,
  `go -C cli test ./...`, `c3local check`. Codex's edits are a *proposal*, never ground
  truth; treat its diff like an untrusted PR.
- **For high-stakes work, run it twice across models** — have Codex and a Claude subagent
  do the same task independently, then reconcile the diff. Cross-model disagreement is
  signal.

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [lagz0ne](https://github.com/lagz0ne)
- **Source:** [lagz0ne/c3-skill](https://github.com/lagz0ne/c3-skill)
- **License:** MIT

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/skill-lagz0ne-c3-skill-codex
- Seller: https://agentstack.voostack.com/s/lagz0ne
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.