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 SiluPanda/codex-agent-loop --skill agent-loopgit clone --depth 1 https://github.com/SiluPanda/codex-agent-loopWrote 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/silupanda/codex-agent-loop/agent-loop)<a href="https://agentmods.dev/skills/silupanda/codex-agent-loop/agent-loop"><img src="https://agentmods.dev/badge/skills/silupanda/codex-agent-loop/agent-loop/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/silupanda/codex-agent-loop/agent-loop"><img src="https://agentmods.dev/badge/skills/silupanda/codex-agent-loop/agent-loop.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 | $0.00050 | $0.00566 |
| Opus 5.5 | $0.00020 | $0.00226 |
| Sonnet 5.5 | $0.00010 | $0.00113 |
| Haiku 4.5 | $0.00005 | $0.00057 |
Grade A, and why
agent-loop 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 1mo 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 — 79 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Agent Loop
Use this skill for multi-step coding tasks that need a bounded agent loop instead of a single local step.
When to use
- The user says loop, agent loop, autonomous, keep iterating, or mentions Claude Code.
- The work will likely require repeated inspect/edit/test cycles.
- A turn cap or explicit approval boundary is useful.
Defaults
--max-turns 8--approval-mode on-write--model gpt-5.4--reasoning-effort high
Important:
- Agent Loop is bounded. It should stop when the task looks complete, when it hits its turn/time budget, when it pauses for approval, or when an error occurs.
- In fallback mode without
OPENAI_API_KEY, Agent Loop runs onecodex execsession. That session may finish on the first turn for short read-only tasks. - A stop reason like
completed_via_fallbackmeans the task completed through the fallback backend; it does not mean the run failed.
Shorthand budgets are also supported:
10m= 10 minutes1h= 1 hour5t= 5 turns
Invocation
Prefer invoking the bundled skill directly:
$agent-loop <task>
Or with a shorthand budget:
$agent-loop 10m <task>
Some Codex builds may also surface the plugin command definition. If yours does, the equivalent command is:
/agent-loop:run <task>
Onboarding helpers
--doctorchecks that the plugin is installed and ready in Codex.--demoruns a safe read-only first-run example and suggests next commands.10m,1h, and5twork as shorthand budgets before the task text.
Approval modes in plain English:
on-write= safenever= hands-offalways= review-everything
Approval behavior
on-writeauto-runs read-only shell inspection commands.on-writepauses before write-like shell commands and all patches.- Use
--resume <state-file> --approve-pendingonly after the user explicitly approves the paused action.
Run artifacts
The runner stores logs and resumable state under:
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.
- 1mo ago First seen · 79 lines · 50 tokens per session scan A 87f04fa29df9
agent-loop is a skill published in the GitHub repository SiluPanda/codex-agent-loop (4 stars, last pushed 6mo ago), licensed MIT. It adds 50 tokens to every session and 566 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 2026-08-31.
Other skills, from other repositories
ralphex-adopt
Convert plans from various source formats (OpenSpec, spec-kit, GitHub/GitLab issues with checklists, generic task-lists, free-form markdown) into ralphex-format plans in docs/plans/. Triggers on "ralphex-adopt", "adopt plan", "convert plan to ralphex", "import plan as ralphex".
ralphex
Run ralphex autonomous plan execution with progress monitoring.
loop
Use only when user explicitly invokes $loop.
reviewable-design-doc
A workflow for creating design documents that organize requirements, architecture, alternatives, decisions, and unresolved questions into reviewable HTML.
visual-html-renderer
A shared workflow for turning requested content into a final HTML document that can be checked and previewed in a browser. It chooses a document structure, renders it, validates the result, and returns a temporary preview link.
plan-preview
A planning-only workflow that creates a temporary HTML preview of an agent’s proposed plan just before the plan is shown to the user.