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 skills add jnopareboateng/codex-claude-subagents --skill claude-subagentsgit clone --depth 1 https://github.com/jnopareboateng/codex-claude-subagentsWrote 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/jnopareboateng/codex-claude-subagents/claude-subagents)<a href="https://agentmods.dev/skills/jnopareboateng/codex-claude-subagents/claude-subagents"><img src="https://agentmods.dev/badge/skills/jnopareboateng/codex-claude-subagents/claude-subagents/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/jnopareboateng/codex-claude-subagents/claude-subagents"><img src="https://agentmods.dev/badge/skills/jnopareboateng/codex-claude-subagents/claude-subagents.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 | $0.00057 | $0.02740 |
| Opus 5.5 | $0.00023 | $0.01096 |
| Sonnet 5.5 | $0.00011 | $0.00548 |
| Haiku 4.5 | $0.00006 | $0.00274 |
Grade A, and why
claude-subagents 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 29d 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 — 151 lines — stays where its author put it; the contents beside it link to each section on GitHub.
claude-subagents
Codex leads. A Claude worker does one bounded task non-interactively and returns a final
message. scripts/csa.py (stdlib Python, WSL/Linux only) starts, supervises, reconciles and
records every run. The worker never asks questions: anything not pre-approved is denied.
When to delegate
- Cross-vendor review of Codex's own changes (a worker never approves its own output).
- Bounded implementation in a scope disjoint from what Codex is editing.
- Second opinion on a design, bug, or plan.
- Not: native-app GUI checks (Codex Desktop computer use), anything needing a human approval mid-task, work that needs the web, or work that is faster to do inline. Keep those with Codex.
Invocation (exactly two forms)
WSL shell:
python3 ~/.codex/skills/claude-subagents/scripts/csa.py <verb> --cwd /home/<you>/projects/<repo> ...
Windows Codex Desktop (PowerShell):
wsl.exe -d Ubuntu-22.04 -e python3 /home/<you>/.codex/skills/claude-subagents/scripts/csa.py <verb> --cwd /home/<you>/projects/<repo> ...
-e runs no shell, so nothing is expanded: use absolute Linux paths, never ~ or $VAR.
--cwd, --prompt and --schema also accept \\wsl.localhost\Ubuntu-22.04\... paths (one or
two leading backslashes). Write each prompt as UTF-8 to <repo>/.agent-runs/prompts/<task>.md
(git-ignored) and pass that path; --prompt - reads stdin (WSL shells). JSON output is ASCII-only.
A --cwd on /mnt/<drive>/ or C:\ is refused unless --allow-windows-fs.
Verbs
| Verb | Use |
|---|---|
start --task ID --prompt FILE|- [--mode review|write] [--scope REL]... [--allow-bash PREFIX]... [--worktree] [--schema FILE|none] [--browser] [--model M] [--effort E] [--max-turns N] [--budget-usd X] [--timeout-min M] [--detach] |
New task. ID matches ^[a-z0-9][a-z0-9-]{0,63}$; a used ID is refused (use resume or a new ID). |
resume --task ID --prompt FILE|- [--allow-bash PREFIX]... [--browser] [--model M] [--effort E] [caps] [--detach] |
New attempt in the same Claude session (claude --resume). Terminal tasks only. --allow-bash/--browser are added for this and later attempts; each attempt records its own in run.json. |
status [--task ID] [--wait SEC] [--json] |
Reconciles liveness, then reports. --wait long-polls and returns on the first transition, or at once if nothing is running. |
result --task ID [--json] |
Final message and structured_output of the latest attempt. |
cancel --task ID |
Only the attempt running when called: SIGINT to its process group, SIGTERM after 15 s, SIGKILL 10 s later; status cancelled. |
list [--active] [--json] |
All tasks of the repo. |
doctor |
Checks every flag and option value the launcher emits against the installed claude (no API calls; permission-rule strings are not validated). |
What ships with it
2 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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.
- 29d ago Changed · -48 lines · +27 tokens per session 58adc987f15e
- 1mo ago Changed · +22 lines ed1b52d9725a
- 1mo ago Changed · +10 lines 22d757820c12
- 1mo ago Changed · +98 lines 42d124367a54
- 2mo ago First seen · 69 lines · 30 tokens per session scan A 9deb4c9b0c62
claude-subagents is a skill published in the GitHub repository jnopareboateng/codex-claude-subagents (15 stars, last pushed 1mo ago), licensed MIT. It adds 57 tokens to every session and 2,740 once invoked, about $0.0002 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 2026-08-30.
claude-subagents compared with skills from other repositories
| Mod | Type | Stars | Last push | Licence | Scan | Tokens | Works with |
|---|---|---|---|---|---|---|---|
| claude-subagentsjnopareboateng/codex-claude-subagents | Skill | 15 | 1mo ago | MIT | A | 57 | Codex |
| skill-routerhussi9/skill-router | Skill | 31 | 2d ago | MIT | A | 69 | Claude Code |
| psy-exp-reviewersoupandpsy/amazing-psycoder-skills | Skill | 38 | 2mo ago | MIT | A | 112 | Codex |
| amazing-psycodersoupandpsy/amazing-psycoder-skills | Skill | 38 | 2mo ago | MIT | A | 110 | Codex |
| psy-ana-designersoupandpsy/amazing-psycoder-skills | Skill | 38 | 2mo ago | MIT | A | 105 | Codex |
| psy-exp-codersoupandpsy/amazing-psycoder-skills | Skill | 38 | 2mo ago | MIT | A | 102 | Codex |
| psy-ana-codersoupandpsy/amazing-psycoder-skills | Skill | 38 | 2mo ago | MIT | A | 94 | Codex |