Ask codex
Use when a Claude Code agent needs to delegate a question, subtask, or review to a local Codex instance.From its SKILL.md
npx -y skills add yeaight7/agent-powerups --skill ask-codexAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- 6 stars6 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.
- runs commandsInstructs the agent to run 4 commands, including `codex --version` and 3 more.
SKILL.md
4.1 KB, ~1.0k tokens by cl100k_base, as published. Nobody here has run it
Ask Codex
Purpose
Use the local Codex CLI as an external worker or advisor from inside a Claude Code session. Hand off a focused task, collect the output, and save a reviewable Markdown artifact before acting on the result.
When to Use
- Need a Codex second opinion on a design decision or code change.
- Want to delegate a self-contained subtask to Codex and inspect the result.
- Exploring a problem from a different model's perspective.
- Do not use when
codexCLI is missing or unauthenticated. - Do not use for tasks that require interactive back-and-forth — Codex output is captured once.
Requirements
codexCLI available in PATH and authenticated.- Verify with:
codex --version
If codex is missing, stop and ask the user to install it. Installation reference:
npm install -g @openai/codex
Inputs
- A specific, scoped prompt or task description.
- Optional: file paths or code snippets to include inline in the prompt.
- Optional: artifact output directory (defaults to
.apx/artifacts/).
Workflow
1. Check the CLI is available
codex --version
If this fails, stop. Do not proceed without codex in PATH.
2. Compose a focused prompt
Keep it narrow. Codex works best with:
- A single question or task.
- Concrete context (file names, function names, error messages).
- A defined output shape ("respond with a patch", "list edge cases", "answer yes/no and explain").
Example prompt:
Review this TypeScript function for correctness. Return a bullet list of issues found.
<paste function here>
3. Run Codex and capture output
codex "Review this TypeScript function for correctness. Return a bullet list of issues found.\n\n$(cat src/cli/commands/validate.ts)"
Redirect output to an artifact file:
codex "Your prompt here" > .apx/artifacts/codex-review-$(date -u +%Y%m%dT%H%M%SZ).md
On Windows (PowerShell):
codex "Your prompt here" | Out-File .apx/artifacts/codex-review-$(Get-Date -Format 'yyyyMMddTHHmmssZ').md
4. Read and evaluate the artifact
Open the artifact before acting on any recommendation. Codex output is a suggestion, not a directive. Apply judgment before implementing.
5. Report
In the response to the user, include:
- The prompt sent to Codex.
- The artifact path.
- A summary of relevant findings.
- Any recommendation you are or are not acting on, with rationale.
Output
Artifact path convention:
.apx/artifacts/codex-<slug>-<timestamp>.md
Artifact should contain:
- The original task or question.
- The exact prompt sent to Codex.
- Raw Codex output.
- Your concise summary.
- Action items or explicit "no action" decision.
Failure Modes
- Missing CLI: stop, ask user to install
codex, do not improvise. - Empty output: report it; do not fabricate a result.
- Off-topic output: note it in the artifact, do not use it.
- Auth failure: report the stderr; check
codex --versionand reauthenticate. - Prompt too broad: narrow the scope before retrying.
apx Integration
apx check ask-codex verifies whether codex is present in PATH.
Prefer the flat APX command after reading this skill:
apx ask-codex "Review this function for edge cases"
apx ask-codex "Return OK only" --json
Nested APX form remains supported:
apx ask codex "Review this function for edge cases"
APX passes the prompt as a positional argument (codex "prompt", not codex -p "prompt"). This matches the Codex CLI's invocation convention. Verify the output mode with codex --help if the captured stdout is empty — some Codex versions open a TUI by default and may require a flag such as --full-auto or -q for non-interactive output. If that applies, invoke codex directly per the workflow above and redirect output manually.
Verification
apx check ask-codex
apx ask-codex "Return OK only" --json
Confirm JSON includes provider, artifactPath, promptLength, and raw output in stdout. Open the artifact before applying advice.
What ships with it
Read from the repository
Just SKILL.md. No reference files, no scripts.
Gives 0 of the 12 instructions most context ai engineering skills give in ~1.0k 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
- compose a narrow and focused prompt
- include concrete context in the prompt
- redirect codex output to a markdown artifact
- evaluate the artifact before acting on recommendations
- summarize findings and rationale in the final response
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.