Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/okisdev/claude-code-fusion/codex-cli-runtimenpx skills add okisdev/claude-code-fusion --skill codex-cli-runtimegit clone --depth 1 https://github.com/okisdev/claude-code-fusionWhat 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 | $0.00015 | $0.01402 |
| Opus 5 | $0.00008 | $0.00701 |
| Sonnet 5 | $0.00003 | $0.00280 |
| Haiku 4.5 | $0.00002 | $0.00140 |
Grade A, and why
codex-cli-runtime scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 48 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Codex runtime
Primary helper:
node "${CLAUDE_PLUGIN_ROOT}/scripts/codex-companion.mjs" <subcommand> [flags]
Subcommand surface:
task [--prompt-file <path>] [--output-schema <path>] [--write] [--background] [--resume <thread-id>|--resume-last|--fresh] [--model <id>] [--effort <level>] [--web] [--network] [--cwd <dir>] [--json] [--] [prompt]review [--scope <auto|working-tree|branch>] [--base <ref>] [--focus <text>] [--background] [--model <id>] [--effort <level>] [--cwd <dir>] [--json]adversarial-review [--scope <auto|working-tree|branch>] [--base <ref>] [--focus <text>] [--background] [--model <id>] [--effort <level>] [--cwd <dir>] [--json]status [job-id] [--all] [--cwd <dir>] [--json]history [--json]result <job-id> [--wait] [--wait-timeout-ms <ms>] [--cwd <dir>] [--json]cancel <job-id> [--cwd <dir>] [--json]setup [--cwd <dir>] [--json]
Execution rules:
-
Programmatic orchestrators pass the complete opaque raw request through stdin with
node "${CLAUDE_PLUGIN_ROOT}/scripts/codex-companion.mjs" task --request-stdin. A write default usestask --transport-default-write --request-stdin. The fixed argv contains no request bytes, and stdin must close after the complete request is written. The same ingress option works with review, status, result, cancel, history, and setup. Slash commands whose Bash tool has no stdin field keep using the private token staging compatibility path:transport-create, one Read of the allocated empty file, Write to that same file, then a fixed--raw-args-tokeninvocation. Never delete, rename, recreate, or change the permissions of the transport file; usetransport-discardwhen Read or Write fails or when the allocated file is not empty. -
Invoke the helper only through foreground
Bashwithtimeout: 600000. -
Place every task option before the first prompt token. Use
--before a prompt that begins with an option shaped token. -
Task, review, and adversarial review reject an implicit working directory below its repository top level before job creation. Pass
--cwdwith either the repository root or the intended subdirectory to choose the sandbox root explicitly. -
The companion forwards
--skip-git-repo-checkto Codex exec. For a task with no ancestor.gitentry, a read only consult automatically receives that flag because it cannot write. A write task fails before job creation unless the request explicitly includes it. The projects trust map inconfig.tomlis not consulted by exec. A gate failure names--skip-git-repo-checkas the remedy. -
Never use Bash background mode. Complexity, duration, and model choice never justify implicit background execution.
-
Pass
--backgroundonly when the received request explicitly contains it. The companion owns detachment and returns a durable job receipt. -
A direct slash command invocation returns that receipt without automatic collection. Direct users inspect progress through status and collect the deliverable through result; when Fusion is installed, its monitor can notify them of completion. Fusion orchestration separately owns one same turn collection attempt capped at 540000ms for jobs it creates. A timeout remains explicitly uncollected.
-
A task is read only unless
--writeis present. Review commands are always read only. -
Leave
--modeland--effortunset unless explicitly requested so Codex configuration remains authoritative. -
Use
--resume <thread-id>only with a real thread identifier returned by Codex. Use--resume-lastfor the newest eligible task thread launched by the current Claude session, or the newest eligible workspace task when no Claude session id is available. -
A foreground task failure of
timeout,policy, orpatch_thrashwith a resumable thread is salvaged by the wrapper's single scripted wind down resume; the resumed job links throughrequest.resumeThreadId, and a second failure terminalizes the package. -
A foreground
--write --model gpt-5.6-soltask emits a nonblocking warning before execution and records it as a diagnostic. With at least four samples, it reports the measured timeout rate and counts for the preceding seven days; otherwise it states the 570s flight budget, recommendinggpt-5.6-terrafor quick shapes andgpt-5.6-lunafor volume shapes. -
Use
--freshonly when the caller explicitly requests a new thread. It cannot be combined with either resume form. -
Use
--webonly when explicitly requested. Use--networkonly with--writeand only when explicitly requested. -
Use
--output-schema <path>only for task mode. The companion resolves the path, requires a regular JSON file at most 256 KiB, and records one JSON parsing result without retrying the task. Nativereviewignores output schemas on tested CLI versions. Review shaped task briefs can use${CLAUDE_PLUGIN_ROOT}/schemas/adversarial-review-verdict.schema.json; adversarial review uses that schema automatically. -
result --waitperforms a bounded wait. When the wait budget expires while the job is still active, it leaves the job unchanged and returns output ending injob: <id>andstate: running. -
Terminal output ends with the Codex thread identifier when available, the job identifier, and
state: done,state: error, orstate: cancelled. Error and cancelled outcomes also include a failure kind. -
Every companion invocation disables Codex
multi_agentandmulti_agent_v2features. A collaboration tool event fails the job even if an upstream configuration bypasses those flags. -
historyreads only the current canonical state root and exposes thread, transport, delivery, semantic, resolved model, and resolved effort fields. Exec sessions remain persisted and resumable locally, but Codex Desktop does not guarantee sidebar visibility for them. -
Return helper stdout exactly as received. Do not interpret it or replace a failed run with Claude-side work.
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.
- yesterday First seen · 48 lines · 15 tokens per session scan A e6b88afdca7e
codex-cli-runtime is a skill published in the GitHub repository okisdev/claude-code-fusion (3 stars, last pushed 14d ago), licensed MIT. It adds 15 tokens to every session and 1,402 once invoked, about $0.0001 per session on Opus 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
agent-framework-py-release
Use when cutting a Python release for the microsoft/agent-framework monorepo. Triggers on "bump py versions", "cut a python release", "prepare release PR for python", "release py packages", "bump python to X.Y.Z", or similar requests to bump Python package versions and prepare a release PR. Handles all four lifecycle…
python-package-management
Guide for managing packages in the Agent Framework Python monorepo, including creating new connector packages, versioning, and the lazy-loading pattern. Use this when adding, modifying, or releasing packages.
foundry-hosted-agent-validation
Step-by-step process for validating a Python Foundry hosted agent sample (under python/samples/04-hosting/foundry-hosted-agents/) end to end — running it locally (native runtime and azd ai agent run) and after deploying it to an Azure AI Foundry project with azd. Use this when asked to validate a hosted agent sample.
build-and-test
How to build and test .NET projects in the Agent Framework repository. Use this when verifying or testing changes.
python-feature-lifecycle
Guidance for package and feature lifecycle in the Agent Framework Python codebase, including stage meanings, feature-stage decorators, feature enums, and how to move APIs from one stage to the next.
python-code-quality
Code quality checks, linting, formatting, and type checking commands for the Agent Framework Python codebase. Use this when running checks, fixing lint errors, or troubleshooting CI failures.