Borrowing it
Nothing to install: this file belongs to yogasw/wick. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/yogasw/wick/master/.claude/skills/agent-resource-guard/SKILL.mdgit clone --depth 1 https://github.com/yogasw/wickWrote 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/yogasw/wick/agent-resource-guard)<a href="https://agentmods.dev/skills/yogasw/wick/agent-resource-guard"><img src="https://agentmods.dev/badge/skills/yogasw/wick/agent-resource-guard/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/yogasw/wick/agent-resource-guard"><img src="https://agentmods.dev/badge/skills/yogasw/wick/agent-resource-guard.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
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 | $0.00231 | $0.09872 |
| Opus 5.5 | $0.00092 | $0.03949 |
| Sonnet 5.5 | $0.00046 | $0.01974 |
| Haiku 4.5 | $0.00023 | $0.00987 |
Grade A, and why
agent-resource-guard 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 25d 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 — 751 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Agent Resource Guard
Wick spawns agents (claude, codex, gemini) as child processes, and those agents spawn their own children — MCP servers, tool subprocesses, browsers. This subsystem bounds what all of that may consume, and measures what it actually does consume.
It is not only memory. Memory is the part that kills, so it gets the deepest treatment; CPU, block-IO, and process count are bounded too, and all four are measured. The name "memory guard" survives in config keys for compatibility — read it as "resource guard".
The one idea to keep
The kernel enforces. Wick observes and explains.
Every design decision below follows from that. A userspace sampler cannot win against a browser that allocates a gigabyte between two ticks, so wick never tries to kill on a sample. It asks the kernel to enforce a ceiling, then reads back what happened so a human gets a sentence instead of exit code 137.
Easy to read backwards
These four have already been misread — twice from the skill's own description, before it said this. If you are answering a question about any of them, quote this section rather than paraphrasing:
| Sounds like | Actually means |
|---|---|
| "measure writes nothing" | Writes no limits. Records usage in full — that is the mode's purpose. |
| "swap must stay off" | Not about OOM-killer predictability. With swap, the process never dies; it thrashes and holds its slot forever. |
| "the guard ships off by default" | Enforcement is off. Usage history is on and independent of the mode. |
| "unavailable without systemd" | Limits still apply without systemd — the cgroupfs backend enforces them. What systemd adds is the ability to name an OOM kill. And measurement, admission, the Resources page, and oom_score_adj never needed either. See the degradation table. |
The shared shape: measurement and enforcement are separate axes. Nearly every misreading here comes from collapsing them into one.
Why this exists
Production wick was being OOM-killed. Not from a leak in wick — an agent or a tool subprocess would balloon (usually a browser or a large file read), and because every one of them was a child of wick inside wick's own cgroup, the kernel's OOM killer took wick down with them.
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.
- 25d ago Changed · +13 lines d8cf5ffb3145
- 2mo ago First seen · 738 lines · 231 tokens per session scan A 88149f41b61d
agent-resource-guard is a skill published in the GitHub repository yogasw/wick (3 stars, last pushed today), licensed MIT. It adds 231 tokens to every session and 9,872 once invoked, about $0.0009 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
n8n
Use when the user wants to build, debug, or extend an n8n workflow - generating workflow JSON from a description, scaffolding a custom TypeScript node, building an AI agent (LangChain cluster), iterating over items, writing Code-node JS, linting an existing workflow, diagnosing a failed execution, or driving a live…
omh-code-review
This is an OMH code-review workflow skill, projected for Agent Skills hosts (Claude Code, Codex, Cursor, opencode, OpenClaw, pi).
redteam-web-detail-pack
Routing and boundary guidance for authorized general web application security testing. Use as a web testing router when the attack surface should be dispatched to more specific web vulnerability skills.
android-pentest
A guide for authorized security testing of Android apps, covering APK inspection, runtime testing, traffic capture, code review, and function hooking. An APK is the installable package used by an Android app.
workflow-builder
Load before calling workflowbuilderbuildworkflow. Use for single-workflow work: new one-off workflows, existing-workflow edits, verification repairs, and workflow-local data tables. For any other new workflow, load intent-recognition first when it is listed. Write or edit a workspace source file, then call…
no-code-app
Use when building an app on a no-code/low-code platform (Bubble, Softr, Glide over Airtable or Xano) — picking the platform, modeling the data, keeping usage-metered cost predictable, and planning the exit before lock-in. NOT a hand-written coded app (that is nextjs), NOT wiring SaaS tools together with no app UI…