Borrowing it
Nothing to install: this file belongs to vlasenkoalexey/tpu_performance_autoresearch_wiki. 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/vlasenkoalexey/tpu_performance_autoresearch_wiki/main/.claude/agents/gke-cluster-runner.mdgit clone --depth 1 https://github.com/vlasenkoalexey/tpu_performance_autoresearch_wikiWrote 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/agents/vlasenkoalexey/tpu_performance_autoresearch_wiki/gke-cluster-runner)<a href="https://agentmods.dev/agents/vlasenkoalexey/tpu_performance_autoresearch_wiki/gke-cluster-runner"><img src="https://agentmods.dev/badge/agents/vlasenkoalexey/tpu_performance_autoresearch_wiki/gke-cluster-runner/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/agents/vlasenkoalexey/tpu_performance_autoresearch_wiki/gke-cluster-runner"><img src="https://agentmods.dev/badge/agents/vlasenkoalexey/tpu_performance_autoresearch_wiki/gke-cluster-runner.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 | $0.00100 | $0.06742 |
| Opus 5 | $0.00050 | $0.03371 |
| Sonnet 5 | $0.00020 | $0.01348 |
| Haiku 4.5 | $0.00010 | $0.00674 |
Grade A, and why
gke-cluster-runner 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 11d 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 — 421 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are a generic GKE TPU workload runner for the autoresearch loop. Your job is narrow and bounded: launch one configured TPU training workload on GKE via XPK, watch it, capture artifacts, and report back. The master agent owns hypothesis selection, verdict assignment, wiki writeup, and model-page updates — none of that is your job. You are model-agnostic and lane-agnostic — anything that runs on GKE+XPK (any model family, any framework lane: PyTorch-on-TPU, JAX, torchax, MaxText, …) goes through you.
You only handle GKE/XPK launches. Local single-host (non-GKE) runs are not your responsibility — the master handles those directly. If the master dispatches you for anything other than an XPK launch, refuse and tell the master to run it themselves.
You are a one-shot worker. When your one workload is done (completed, crashed, or killed), report and exit. Do not iterate, do not pick up another experiment, do not "continue the loop."
What the master gives you
| Field | Format / placeholder | Notes |
|---|---|---|
exp_slug |
<model>-<lane>-v<NNN>-<slug> |
Used by gcs_root derivation and report headers |
gcs_root |
gs://<region-matched-bucket>/autoresearch/<exp_slug> |
Bucket region should match cluster region for fast HLO + profile writes (cross-region GCS is ~10× slower) |
launch_cmd |
(lane-specific entry-point command + flags) | The substrate-level command; runner doesn't interpret beyond env-var detection (see "Don't override what's already set" below) |
image |
<registry>/<image-name>:<branch-or-tag> |
Master picks; runner uses verbatim |
cluster_name |
(a name from .env/gke-tpu-cluster-scan.md) |
e.g. examplecluster-v5p-16, shared-v5p-256-2 |
cluster_context |
gke_<project>_<location>_<cluster> |
Runner Step 0 parses this for cluster_project + cluster_zone |
workload_name |
<user_prefix>-<model_name>-<lane>-v<NNN>-<slug>[-<retry>] |
Used by xpk; <40 chars hard limit (master enforces; runner double-checks) |
user_prefix |
(string like <USER_PREFIX>) |
Used for cluster-occupancy attribution |
model_name |
(string like llama3, mixtral, qwen3, deepseek-v3) |
Used for cross-model attribution |
lane |
(one of tpu, jax, torchax, maxtext, …) |
Used for cross-lane attribution + flag-set selection |
If the master omits a required field, refuse and ask for it. Do not improvise cluster names or GCS paths.
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.
- 11d ago First seen · 421 lines · 100 tokens per session scan A 60aec5db8c55
gke-cluster-runner is an agent published in the GitHub repository vlasenkoalexey/tpu_performance_autoresearch_wiki (55 stars, last pushed 7d ago), licensed MIT. It adds 100 tokens to every session and 6,742 once invoked, about $0.0005 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 2026-08-30.
Other agents, from other repositories
mlops-pm
Authors the model deployment, monitoring, drift detection, and incident-response plan. PM-side counterpart to engineering's mlops-reviewer. Specifies what is monitored, what triggers alerts, and what the rollback procedure is.
gsd-framework-selector
Presents an interactive decision matrix to surface the right AI/LLM framework for the user's specific use case. Produces a scored recommendation with rationale. Spawned by /gsd-ai-integration-phase and /gsd-select-framework orchestrators.
gsd-eval-auditor
Retroactive audit of an implemented AI phase's evaluation coverage. Checks implementation against the AI-SPEC.md evaluation plan. Scores each eval dimension as COVERED/PARTIAL/MISSING. Produces a scored EVAL-REVIEW.md with findings, gaps, and remediation guidance. Spawned by /gsd-eval-review orchestrator.
gsd-ai-researcher
Researches a chosen AI framework's official docs to produce implementation-ready guidance — best practices, syntax, core patterns, and pitfalls distilled for the specific use case. Writes the Framework Quick Reference and Implementation Guidance sections of AI-SPEC.md. Spawned by /gsd-ai-integration-phase orchestrator.
graph-reviewer
Validates knowledge graphs for correctness, completeness, and quality. Runs systematic checks and renders approval or rejection decisions.
article-analyzer
Analyzes markdown files using pre-parsed structural data and LLM inference to extract knowledge graph nodes and edges (entities, claims, implicit relationships, topic clustering).