Getting it into your agent
This one installs as part of its plugin. Adding the marketplace and installing the plugin brings it with everything else the plugin ships.
/plugin marketplace add vsladkov/claudex-stereo/plugin install stereoWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/vsladkov/claudex-stereo/codex-cli-runtime)<a href="https://agentmods.dev/skills/vsladkov/claudex-stereo/codex-cli-runtime"><img src="https://agentmods.dev/badge/skills/vsladkov/claudex-stereo/codex-cli-runtime/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/vsladkov/claudex-stereo/codex-cli-runtime"><img src="https://agentmods.dev/badge/skills/vsladkov/claudex-stereo/codex-cli-runtime.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00033 | $0.01148 |
| Opus 5.5 | $0.00013 | $0.00459 |
| Sonnet 5.5 | $0.00007 | $0.00230 |
| Haiku 4.5 | $0.00003 | $0.00115 |
Grade A, and why
codex-cli-runtime scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 21d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 68 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Codex Runtime
Use this skill only inside the stereo:codex-rescue subagent.
Primary helper — the flags as separate arguments, the task text on stdin through a quoted heredoc:
node "${CLAUDE_PLUGIN_ROOT}/scripts/codex-companion.ts" task <taskFlags> <<'STEREO_EOF'
<task text>
STEREO_EOF
<taskFlags>is each routing flag as its own argument, in this order when present:--write,--background,--resume(or--resume-last) or--fresh,--model '<selection>', and--effort '<effort>', every value single-quoted (an embedded'becomes'"'"'). Nothing else goes on the command line: never put the task text, or any other user text, in an argument or a shell string.- The task text — the user's request with the routing flags removed, or the prompt the
codex-promptingskill tightened from it — goes between the heredoc lines verbatim. The quoted delimiter'STEREO_EOF'keeps the shell from expanding anything inside it. If a line of the text is exactlySTEREO_EOF, append digits to the delimiter (STEREO_EOF_2) on both lines until no line of the text matches it. With--resumeand no new text, omit the heredoc. - Run the Bash call with a
timeoutof 600000 ms (the tool's maximum): a foreground rescue task routinely outlasts the default two minutes.
Execution rules:
- The rescue subagent is a forwarder, not an orchestrator. Its only job is to invoke
taskonce and return that stdout unchanged. - Call only the helper: no hand-rolled
git, no direct Codex CLI strings, no other Bash activity. - Do not call
setup,review,adversarial-review,status,result, orcancelfromstereo:codex-rescue. - Use
taskfor every rescue request, including diagnosis, planning, research, and explicit fix requests. - You may use the
codex-promptingskill to rewrite the user's request into a tighter Codex prompt before the singletaskcall. - That prompt drafting is the only Claude-side work allowed. Do not inspect the repo, solve the task yourself, or add independent analysis outside the forwarded prompt text.
- Leave
--effortunset unless the user explicitly requests a specific effort, and then forward the value verbatim. Never invent an effort the user did not ask for. - Leave model unset by default. Add
--modelonly when the user explicitly asks for one, then pass that value through verbatim, with or without itscodex:prefix. - The companion resolves and validates the model and the effort itself. Never resolve, expand, or check either value; a value the companion refuses comes back as the failure line below.
- Default to a write-capable Codex run by adding
--writeunless the user explicitly asks for read-only behavior or only wants review, diagnosis, or research without edits.
Command selection:
- Use exactly one
taskinvocation per rescue handoff. - The Agent invocation and Bash call are always foreground. Strip
--waitand make a foregroundtaskcall; map--backgroundtotask --background. Neither flag is part of the natural-language task text. - If the forwarded request includes
--model, pass its value totaskas--model '<selection>'without expanding aliases. - If the forwarded request includes
--effort, pass its value totaskas--effort '<effort>'. - Forward
--resumeand--freshtotaskunchanged as flags, never as task text; the CLI accepts both directly (--resumeis an alias of--resume-last).--resumealways resumes, and--freshalways runs fresh, even when the request text is ambiguous. --effort: the values arenone,minimal,low,medium,high,xhigh,max,ultra.task --resume-last: internal helper for "keep going", "resume", "apply the top fix", or "dig deeper" after a previous rescue run; it continues this session's latest rescue thread.
Safety rules:
- Forward the user's task text unchanged apart from the routing flags, unless you tighten it with the
codex-promptingskill; a tightened prompt keeps the user's meaning and scope. - Do not inspect the repository, read files, grep, monitor progress, poll status, fetch results, cancel jobs, summarize output, or do any follow-up work of your own.
- Return the stdout of the
taskcommand exactly as-is. - If the Bash call fails or Codex cannot be invoked, return exactly
Codex rescue failed: <first line of the error>. Run /stereo:setup to check the Codex CLI.and add nothing else.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 21d ago Changed · +19 lines · +14 tokens per session 47ad86c37d2c
- 1mo ago Changed 223a448c9b87
- 1mo ago First seen · 49 lines · 19 tokens per session scan A b74209fa4a85
codex-cli-runtime is a skill published in the GitHub repository vsladkov/claudex-stereo (4 stars, last pushed 21d ago), licensed Apache-2.0. It adds 33 tokens to every session and 1,148 once invoked, about $0.0001 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
goal-test
A local experiment for testing a goal command that keeps an AI coding session working until a stated condition is judged complete. It uses a separate language model to evaluate the conversation after each assistant turn.
loop-gh-pr
A recipe for a scheduled GitHub pull-request watcher. It checks open pull requests for new activity and prepares read-only review drafts without posting comments or merging anything.
goal-plan-clean
A guide for creating a reusable plan-review loop. It repeatedly checks a plan, fixes blocking issues and suggestions, and stops after approval or a small number of review rounds.
loop-bb-pr
A command that prepares a repeating check for a Bitbucket pull request. A pull request is a request to review and merge code changes into a project.
loop-test
A local experiment for designing repeating checks with Claude Code’s loop command. It covers both fixed intervals and checks that schedule themselves again after each run.
commit
A guided process for committing code changes to Git. It checks the working tree, follows a conventional commit format, links a task when available, reviews the changes, and asks for approval.