tpu_performance_autoresearch_wiki: Skill for Claude Code

.claude/skills/edit-model-code/SKILL.md

edit-model-code is a skill for Claude Code from vlasenkoalexey/tpu_performance_autoresearch_wiki. It costs 146 tokens per session (2,709 once invoked), scanned A, original, MIT.

A set of guidelines for making small, controlled changes to model code in an isolated experiment copy. The autoresearch loop is a repeated process of testing model changes and measuring their results.

In plain words
What is it for?
Use it before editing model files in an autoresearch experiment, especially when considering a change that could spread across several parts of the code.
Why use it?
It helps prevent accidental behavior changes and keeps each experiment focused on one testable idea.

Skill for Claude Code

Written for Claude Code: installed under .claude/. Also seen: reads .claude/ paths; mentions subagents.

This is vlasenkoalexey/tpu_performance_autoresearch_wiki's own configuration. It tells Claude Code how to work on tpu_performance_autoresearch_wiki itself, so it is not a mod to install elsewhere. Copy it as a starting point and replace the rules that are about this project. Everything tpu_performance_autoresearch_wiki configures →

Reuse

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.

Copy the file
curl -O https://raw.githubusercontent.com/vlasenkoalexey/tpu_performance_autoresearch_wiki/main/.claude/skills/edit-model-code/SKILL.md
Clone the repo
git clone --depth 1 https://github.com/vlasenkoalexey/tpu_performance_autoresearch_wiki

Made for: Claude Code.

Wrote 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.

agentmods badge for edit-model-code

README.md
[![agentmods](https://agentmods.dev/badge/skills/vlasenkoalexey/tpu_performance_autoresearch_wiki/edit-model-code.svg)](https://agentmods.dev/skills/vlasenkoalexey/tpu_performance_autoresearch_wiki/edit-model-code)
Your own site
<a href="https://agentmods.dev/skills/vlasenkoalexey/tpu_performance_autoresearch_wiki/edit-model-code"><img src="https://agentmods.dev/badge/skills/vlasenkoalexey/tpu_performance_autoresearch_wiki/edit-model-code.svg" alt="Measured on agentmods" height="20"></a>
Per session 146 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 2,709 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe. Third-party audits
  • NVIDIA SkillSpector warn 7 Sept 2026
SkillSpector: 4 findings, up to medium

These are SkillSpector’s own severities. On a checked sample its high-severity flags on skills were ~96% false positives — a documented command, a public API, a “never do X” rule — so we show them as a caution to read, not a verdict. Why →

  • medium Agent Snooping · line 8
    Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.
    Fix: Remove all code or instructions that list or read other skills' files or directories. Skills should operate independently; cross-skill access is a privilege escalation.
  • medium Agent Snooping · line 148
    Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.
    Fix: Remove all code or instructions that list or read other skills' files or directories. Skills should operate independently; cross-skill access is a privilege escalation.
  • medium MCP Rug Pull · line 115
    Docker image references without a specific tag (:latest is implicit) or digest (@sha256:...) can be silently replaced by a malicious image.
    Fix: Pin the image: image:tag or image@sha256:abc123
  • medium Agent Snooping · line 150
    Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.
    Fix: Remove all code or instructions that list or read other skills' files or directories. Skills should operate independently; cross-skill access is a privilege escalation.
How audits are shown
Origin original No closer match found in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5.1 $0.00146 $0.02709
Opus 5 $0.00073 $0.01354
Sonnet 5 $0.00029 $0.00542
Haiku 4.5 $0.00015 $0.00271

Measured 8d ago against content hash 576a5ffaeef0, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-08, from the pricing page.

Security

Grade A, and why

edit-model-code 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 8d 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.

.claude/skills/edit-model-code/SKILL.md · 154 lines

How it starts

The opening of the file, as written. The whole thing — 154 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are about to modify model code inside a per-experiment fork. Read these guidelines FIRST and keep them in scope while editing. The goal is one minimal, surgical change that tests exactly one hypothesis.

This skill adapts Andrej Karpathy's LLM-coding guidelines (MIT-licensed) to the autoresearch loop. Where the upstream conflicts with this project's invariants, the autoresearch version wins.

When to invoke this skill

  • BEFORE step 4 ("Implement") of program.md's experiment loop — before opening any file in wiki/experiments/<model>_autoresearch_optimization/<lane>/.repo/<exp-name>/<model-repo>/.
  • Re-invoke if the edit scope grows mid-session (you start touching files outside the original plan).
  • Re-invoke for any cross-cutting refactor you're tempted to do "while you're in there".

NOT for: wiki page writes, infra scripts, the build step, the launch step. Those have their own discipline elsewhere.

The four principles (Karpathy, adapted)

1. Think before coding — and document, don't ask

Surface assumptions in the experiment page, not by interrupting the loop.

The gke-cluster-runner is autonomous; the master loop can't pause to interactively clarify with the user. So the adaptation is:

  • State your assumptions in the experiment page's "Hypothesis under test" or "Setup" section before editing.
  • If multiple interpretations of the hypothesis exist, pick the smaller-blast-radius one and document why in the experiment page. Don't make both changes.
  • If something is genuinely unclear and the wrong choice would burn cluster time, refute the hypothesis cheaply with an inconclusive run and surface the ambiguity in the verdict — that's a valid outcome of the loop.

2. Simplicity first — one mechanism per experiment

Minimum code that tests this one hypothesis. Nothing speculative.

  • No features beyond what the hypothesis tests.
  • No "while I'm here" cleanups, no abstractions for a single-use code path.
  • No error handling for impossible scenarios (the training job will surface real errors).
  • If you find yourself writing 200 lines of plumbing to test a 5-line idea, the hypothesis is wrong — refute it cheaply or reformulate.
  • One mechanism per experiment. Don't bundle "fix imports + tune block size + add a fallback path". The wiki's signal depends on one variable changing per run.

Read the full file on GitHub · 154 lines

Changes

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.

  1. 8d ago First seen · 154 lines · 146 tokens per session scan A 576a5ffaeef0

Subscribe to this mod's changes

edit-model-code is a skill published in the GitHub repository vlasenkoalexey/tpu_performance_autoresearch_wiki (54 stars, last pushed 4d ago), licensed MIT. It adds 146 tokens to every session and 2,709 once invoked, about $0.0007 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.