Codex cli agent

Skill ImL1s/agent-cli-skills/skills/codex-cli-agent

Run OpenAI Codex CLI headlessly via codex-exec.sh (codex exec). Use when the user asks for Codex, GPT coding agent, codex exec, Codex subagents, or spawn_agents_on_csv. -r / read-only sandbox blocks workspace writes — use -s workspace-write when the prompt must write files. -T prefers subagents. Orchestrator owns git (--no-git).From its SKILL.md

Install

npx -y skills add ImL1s/agent-cli-skills --skill codex-cli-agent

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

2 stars2 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

2.2 KB, 537 tokens by cl100k_base, as published. Nobody here has run it

Codex CLI (`codex exec`) Executor

When to use

User wants Codex / codex exec as the coding agent
Need Codex subagents, agent TOML profiles, or CSV fan-out
Sandboxed writes (-s workspace-write) vs strict read-only reviews (-r)

Wrapper

./codex-exec.sh -C /path/to/repo -s workspace-write "Implement X. NO git."
./codex-exec.sh -r -C /path/to/repo "Review for bugs"          # sandbox read-only
./codex-exec.sh -T -C /path/to/repo "Parallelize with subagents"
./codex-exec.sh -j -o /tmp/last.txt -f /tmp/bundle.txt
./codex-exec.sh --enable some_feature -m <model> "…"

Flags: -m · -t · -C · -s read-only|workspace-write|danger-full-access · -o last-message file · -f · -j/--json · -r · -T · --skip-git-repo-check (default on) · --require-git-repo · --ephemeral · --add-dir · --image · --enable · --profile · --no-git · -- passthrough.

-C is absolutized before use (avoids nested relative cwd).

Sandbox vs writing answer files

If the prompt must write a file under the workspace, do not use -r / read-only. That fails with writing is blocked by read-only sandbox after the model finishes reasoning. Use -s workspace-write and instruct “only write the designated path”, or capture stdout.

Native multi-agent

Subagents (parallel spawn + synthesize); custom roles in ~/.codex/agents/*.toml or .codex/agents/
Experimental batch: spawn_agents_on_csv (one worker per CSV row)
Feature flags: codex features list|enable|disable
-T asks Codex to prefer subagents / CSV fan-out when useful

Models

Do not hard-code a stale model table. Discover via your Codex build / account docs. Pin with -m when automation requires stability.

Discipline

Always --skip-git-repo-check outside git repos. Exit code untrusted. Orchestrator owns git.

What ships with it: 4 files

16.1 KB alongside SKILL.md, 4 of them executable

lib/

common.shruns6.8 KB
parse_stream_json.pyruns3.5 KB
pty_run.pyruns1.7 KB

codex-exec.shruns4.0 KB

Gives 0 of the 12 instructions most context ai engineering skills give in 537 tokens

Counted across 1,328 of the 2,349 authors here whose files we hold, read 2026-09-06

Dispatch a fresh subagent for each taskin 76 of 1328, across 59 files
Perform spec compliance review before code quality reviewin 44 of 1328, across 34 files
Dispatch a final code reviewer after all tasksin 38 of 1328, across 26 files
Answer subagent questions before allowing implementationin 36 of 1328, across 26 files
Use the least powerful model capable of the taskin 33 of 1328, across 26 files
Create a TodoWrite list for all tasksin 32 of 1328, across 22 files
Perform a task review after each implementationin 31 of 1328, across 24 files
Extract all tasks and context from the planin 29 of 1328, across 20 files
Provide full task text to subagentsin 28 of 1328, across 20 files
Use git worktrees for isolated workspacesin 25 of 1328, across 20 files
Specify the model explicitly when dispatching a subagentin 23 of 1328, across 18 files
Execute all tasks from the plan without stoppingin 21 of 1328, across 16 files

Said here and by no other author read

Use -s workspace-write for file writing tasks
Use -r for read-only reviews
Use -T to prefer subagents
Absolutize the -C path before use
Pin models with -m for stability

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.