steering-block-selection

steering-block-selection is a skill for Claude Code from zjunlp/Mechanist. It costs 249 tokens per session (712 once invoked), scanned A, original, MIT.

Guidance for choosing which internal layers or processing sites of a transformer model to modify during representation experiments. It covers methods such as activation patching, steering, and feature scaling.

In plain words
What is it for?
Use it to screen or select layers, attention components, memory-like residual pathways, or other sites for controlled model-intervention experiments.
Why use it?
Choosing too few sites can hide an effect, while changing too many can create a broad model failure that does not support a specific explanation.

Skill for Claude Code

Written for Claude Code: shipped in a Claude Code plugin.

Part of the mechanist plugin — 54 skills, 4 agents shipped together

Install

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.

agentmods
npx agentmods add skills/zjunlp/mechanist/steering-block-selection
Any agent
npx skills add zjunlp/Mechanist --skill steering-block-selection
Clone the repo
git clone --depth 1 https://github.com/zjunlp/Mechanist

Made for: Claude Code.

Or install mechanist, the plugin that ships this one along with the rest of its 54 skills, 4 agents.

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 steering-block-selection

README.md
[![agentmods](https://agentmods.dev/badge/skills/zjunlp/mechanist/steering-block-selection.svg)](https://agentmods.dev/skills/zjunlp/mechanist/steering-block-selection)
Your own site
<a href="https://agentmods.dev/skills/zjunlp/mechanist/steering-block-selection"><img src="https://agentmods.dev/badge/skills/zjunlp/mechanist/steering-block-selection.svg" alt="Measured on agentmods" height="20"></a>
Per session 249 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 712 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
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 $0.00249 $0.00712
Opus 5 $0.00125 $0.00356
Sonnet 5 $0.00050 $0.00142
Haiku 4.5 $0.00025 $0.00071

Measured 6d ago against content hash 7ece6cf1ac74, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-06, from the pricing page.

Security

Grade A, and why

steering-block-selection 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 6d 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.

skills/experiment-tips/steering-block-selection/SKILL.md · 42 lines

What it actually says

Steering Block / Layer Selection

The point

Where and how many sites you intervene on changes both the effective strength of the intervention and whether a localization claim is testable.

  • Too few (single site on a deep stack) → downstream norm/attention denoise it → looks like "no effect".
  • Too many (whole stack) → off-distribution collapse → the effect is collapse-driven and no longer localized.

Choosing where to intervene

Pick the site(s) by a screening method, by heuristic, or both.

By method (pre-screen the layers):

  • Gradient-based — rank layers by gradient / attribution score w.r.t. the target.
  • Activation-based — rank by an activation signal (probe accuracy, diff-mean magnitude, separability).

By heuristic:

  1. Mid-to-late layers usually work best — start there.
  2. If a single layer shows no effect, intervene on 3–5 layers instead.
  3. To sweep the whole stack, use spaced intervals (e.g. every 2–3 layers) rather than every layer.

What to intervene on (transformer)

Target the attention, the MLP, or the residual stream — choose by what the claim is about.

Note: circuit discovery generally studies attention heads + MLP.

Two rules to keep it honest

  • Match the claim. If the claim is about a region (e.g. "early layers 0–7"), intervene on a window covering that region and use matched windows elsewhere as null controls — a single site can't adjudicate a regional claim. Report the localization you actually observe; don't rewrite the claim to fit the data.
  • Never copy a raw index across models of different depth. "Layer 16 of 32" → scale by relative depth (0.5 → layer 24 of 48), then check neighbors.

Composition

Lock the site set here first (the coefficient plateau is site-dependent), then sweep α via ../steering-coefficient-tuning/. Re-sweep the coefficient whenever the site set changes.

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. 6d ago First seen · 42 lines · 249 tokens per session scan A 7ece6cf1ac74

Subscribe to this mod's changes

steering-block-selection is a skill published in the GitHub repository zjunlp/Mechanist (55 stars, last pushed 10d ago), licensed MIT. It adds 249 tokens to every session and 712 once invoked, about $0.0012 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.