agentleFS
wrg32786wrg32786/aigent-os/skills/codex-adapter/SKILL.md

codex-adapter

Skill25 starsChanged 7 days agoMITView on GitHub

Route one mechanical task to OpenAI Codex CLI with review before merge. Use when explicitly asked to hand Codex a rename, config edit, or scripted replacement.

2 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

2 rules. About 0 are common, 1 is distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

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

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

    Not a fit: architecture decisions, anything requiring judgment about tradeoffs, anything touching credentials/secrets/production config, …

    line 23
  2. Distinctive:

    Not a general multi-LLM router -- it's one task class routed to one executor.

    line 59

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “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 Adapter

This is aigent-OS's first (and currently only) non-Claude executor: a thin wrapper around daemons/codex-adapter.sh that runs a single bounded, mechanical task through the Codex CLI's non-interactive codex exec mode.

Scope -- what "bounded and mechanical" means

Good fits: a scripted rename, a config/gitignore entry, a repetitive multi-file edit with a clear before/after, a lint-fix pass. The task prompt you hand it should be checkable by a human (or you) glancing at the resulting diff -- if you can't describe in one sentence what "done" looks like, it's not bounded enough for this adapter; do it yourself instead.

Not a fit: architecture decisions, anything requiring judgment about tradeoffs, anything touching credentials/secrets/production config, anything the user hasn't explicitly asked to route to Codex.

Before you invoke it

  1. Confirm the Codex CLI is actually available: command -v codex (or whatever AIGENT_CODEX_BIN is set to). If it isn't installed, tell the user instead of guessing at a path.
  2. State the bounded task in one sentence, and the target directory. If either is unclear, ask -- don't invoke on a guess.

Invocation

bash daemons/codex-adapter.sh "<bounded task prompt>" --dir <path> [--sandbox <mode>]
  • AIGENT_CODEX_BIN (env, default codex) -- path or name of the Codex CLI binary. Never hardcode a machine-specific path in the invocation; set the env var if the binary isn't on PATH.
  • --dir <path> -- working directory the task runs in. Default: current directory.
  • --sandbox <mode> -- passed through to codex exec --sandbox. Default workspace-write (Codex may read/write inside --dir, nothing wider).

Example:

AIGENT_CODEX_BIN=codex bash daemons/codex-adapter.sh \
  "Add a .gitignore entry for *.tmp files" --dir .

After it runs -- the review gate

The adapter never commits or pushes. It writes:

  • .aigent/codex-runs/<timestamp>/output.log -- the raw Codex CLI transcript
  • .aigent/codex-runs/<timestamp>/review.diff -- a working-tree diff (tracked changes + a synthetic diff for any new untracked files) if --dir is a git repository
  • .aigent/codex-runs/<timestamp>/status.txt -- git status --porcelain for the same tree

Read review.diff before doing anything else. Treat a Codex-produced change exactly like a sub-agent's diff: verify it does what was asked, then decide whether to stage/commit it yourself. If --dir isn't a git repository, the adapter says so explicitly and you review the files directly instead.

What this is not

Not a general multi-LLM router -- it's one task class routed to one executor. Extending this to route by task type, or to wire additional CLIs (Gemini CLI, opencode, etc.) behind the same interface, is future work, not something to imply is already live.

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.