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 redai-infra/Relax --skill ssh-ray-clustergit clone --depth 1 https://github.com/redai-infra/RelaxWrote 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/redai-infra/relax/ssh-ray-cluster)<a href="https://agentmods.dev/skills/redai-infra/relax/ssh-ray-cluster"><img src="https://agentmods.dev/badge/skills/redai-infra/relax/ssh-ray-cluster/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/redai-infra/relax/ssh-ray-cluster"><img src="https://agentmods.dev/badge/skills/redai-infra/relax/ssh-ray-cluster.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 | $0.00721 |
| Opus 5 | $0.00016 | $0.00360 |
| Sonnet 5 | $0.00007 | $0.00144 |
| Haiku 4.5 | $0.00003 | $0.00072 |
Grade C, and why
ssh-ray-cluster scanned grade C with 1 finding 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 9d 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.
Recursive force deletehighDestructive command
rm -rf with a variable or a broad path is one typo away from removing the wrong tree.
- `rm -rf` on `/tmp/ray/` or session directories How it starts
The opening of the file, as written. The whole thing — 76 lines — stays where its author put it; the contents beside it link to each section on GitHub.
SSH Debug Loop
Three-step cycle: submit -> check logs -> analyze & fix -> repeat.
Prerequisites
Read SSH credentials and RELAX_PROJECT_ROOT from auto-memory (reference_ray_cluster_ssh.md). Ask the user if missing — never hard-code in this file.
Step 1: Submit Task via SSH
Use paramiko to SSH into the cluster, cd to the project root, and execute the user's command.
python3 -c "
import paramiko, shlex
ssh = paramiko.SSHClient()
ssh.set_missing_host_key_policy(paramiko.AutoAddPolicy())
ssh.connect(HOST, port=PORT, username=USER, password=PASS, timeout=10)
cmd = f'cd {shlex.quote(RELAX_PROJECT_ROOT)} && <USER_COMMAND>'
try:
stdin, stdout, stderr = ssh.exec_command(cmd, timeout=60)
print(stdout.read().decode())
err = stderr.read().decode()
if err: print('STDERR:', err)
except Exception: pass # long-running commands may timeout — that's OK
finally: ssh.close()
"
Key rule: All project-relative commands (bash scripts/..., tail log/...) MUST have cd $RELAX_PROJECT_ROOT && in the same command string. Paramiko opens a fresh shell each call.
For backgrounded launches, verify separately:
pgrep -af 'ray-job.sh' | head
ray job list 2>&1 | grep RUNNING | head
Step 2: Check Logs Locally
The log file is on a shared filesystem mounted locally. Read it directly:
# Find the latest log
ls -lt log/<model>-*.log | head -5
# Read the tail for errors
tail -200 log/<run-name>.log
Use the Read tool on the log file path. Search for keywords: Error, Exception, Traceback, FAILED, RuntimeError, AssertionError.
Check frequency: Wait at least 1 minute between log checks. Don't poll more frequently — training jobs take minutes to hours, and frequent checks waste context.
Step 3: Analyze & Fix
- Identify the error from the log (traceback, error message, hang pattern).
- Fix the code if the root cause is clear — edit the source file directly.
- Add debug logging if the root cause is unclear — add targeted
logger.info/logger.errorcalls to narrow down the issue. - Go back to Step 1 — resubmit the task and repeat until resolved.
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.
- 9d ago First seen · 76 lines · 33 tokens per session scan C f7288b6d914d
ssh-ray-cluster is a skill published in the GitHub repository redai-infra/Relax (580 stars, last pushed 11d ago), licensed Apache-2.0. It adds 33 tokens to every session and 721 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it C with 1 finding (recursive force delete). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
verl-rl-training
Provides guidance for training LLMs with reinforcement learning using verl (Volcano Engine RL). Use when implementing RLHF, GRPO, PPO, or other RL algorithms for LLM post-training at scale with flexible infrastructure backends.
openrlhf-training
High-performance RLHF framework with Ray+vLLM acceleration. Use for PPO, GRPO, RLOO, DPO training of large models (7B-70B+). Built on Ray, vLLM, ZeRO-3. 2× faster than DeepSpeedChat with distributed architecture and GPU resource sharing.
verl-rl-training
Provides guidance for training LLMs with reinforcement learning using verl (Volcano Engine RL). Use when implementing RLHF, GRPO, PPO, or other RL algorithms for LLM post-training at scale with flexible infrastructure backends.
openrlhf-training
High-performance RLHF framework with Ray+vLLM acceleration. Use for PPO, GRPO, RLOO, DPO training of large models (7B-70B+). Built on Ray, vLLM, ZeRO-3. 2× faster than DeepSpeedChat with distributed architecture and GPU resource sharing.
verl-rl-training
Provides guidance for training LLMs with reinforcement learning using verl (Volcano Engine RL). Use when implementing RLHF, GRPO, PPO, or other RL algorithms for LLM post-training at scale with flexible infrastructure backends.
openrlhf-training
High-performance RLHF framework with Ray+vLLM acceleration. Use for PPO, GRPO, RLOO, DPO training of large models (7B-70B+). Built on Ray, vLLM, ZeRO-3. 2× faster than DeepSpeedChat with distributed architecture and GPU resource sharing.