instruction-ablation

instruction-ablation is a skill for Claude Code, Cursor from emaraschio/cursor-commands. It costs 78 tokens per session (1,584 once invoked), scanned A, original, MIT.

An evidence-based workflow for reducing a set of instructions to the smallest set that is still useful. It runs a real task with minimal guidance, adds an instruction only after a repeated failure, and keeps it only if verification improves.

In plain words
What is it for?
Use it to simplify agent instructions, test whether a rule is needed, and compare task verification before and after an instruction change.
Why use it?
It helps remove unnecessary prompt rules that add complexity without improving results. Each proposed rule is tested against observed failures rather than added by guesswork.

Skill for Claude CodeCursor

Written for Claude Code and Cursor: disable-model-invocation in frontmatter, but also installed under .cursor/.

Good fit Use it to simplify agent instructions, test whether a rule is needed, and compare task verification before and after an instruction change.

Compare 6 skills from other repositories ↓
Install with agentmods
npx agentmods add skills/emaraschio/cursor-commands/instruction-ablation
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.

Any agent
npx skills add emaraschio/cursor-commands --skill instruction-ablation
Clone the repo
git clone --depth 1 https://github.com/emaraschio/cursor-commands

Made for: Claude Code, Cursor.

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 instruction-ablation

README.md
[![agentmods](https://agentmods.dev/badge/skills/emaraschio/cursor-commands/instruction-ablation.svg)](https://agentmods.dev/skills/emaraschio/cursor-commands/instruction-ablation)
Your own site
<a href="https://agentmods.dev/skills/emaraschio/cursor-commands/instruction-ablation"><img src="https://agentmods.dev/badge/skills/emaraschio/cursor-commands/instruction-ablation.svg" alt="Measured on agentmods" height="20"></a>
Per session 78 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 1,584 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.
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 $0.00078 $0.01584
Opus 5 $0.00039 $0.00792
Sonnet 5 $0.00016 $0.00317
Haiku 4.5 $0.00008 $0.00158

Measured 5d ago against content hash a01fde6ac942, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-08, from the pricing page.

Security

Grade A, and why

instruction-ablation 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 5d 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.

.cursor/skill-contracts/instruction-ablation/SKILL.md · 138 lines

How it starts

The opening of the file, as written. The whole thing — 138 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Instruction ablation

Role

You run an instruction ablation: rebuild a minimal instruction set from a bare baseline on one real task. You execute the task in-session under GUARDRAILS only, record what already works, and propose a new instruction only when the same failure pattern repeats. For each proposal: explain the failure, write the smallest instruction, retest, and keep it only if VERIFICATION improves. You do not rewrite the user's rules, skills, hooks, or memories unless they say apply now with targets.

When to use

Use when the user invokes /instruction-ablation, asks to delete and rebuild AI instructions from evidence, or wants to thin rules/skills/hooks after models improved. For debugging a pasted prompt with a tiny eval suite, use prompt-eval-debug. For composing a new structured prompt from a rough ask, use structure-prompt. For a six-part Goal before autonomous work, use define-agent-goal.

Ablation charter

Start bare. Run a real task. Add an instruction only after repeated failure. Retest. Keep only if verification improves. Never claim the environment was bare without confirmation.

Workflow

Run phases in order.

Phase 0: Intake

  1. Require four fields:
    • TASK: the real work to attempt
    • GUARDRAILS: rules that must not be violated
    • EXIT CRITERIA: exactly what "done" means
    • VERIFICATION: how to test or inspect the result
  2. If any field is missing, ask one focused question per gap. Do not invent values.
  3. Optionally accept a paste or description of the user's current instruction stack (rules, skills, hooks, memories) for later delete-candidate analysis.
  4. If the task is production-destructive and GUARDRAILS lack explicit consent for that blast radius, stop and ask before the bare run.

Phase 1: Baseline gate

A slash command cannot disable Cursor skills, hooks, memories, or user rules.

  1. Ask the user to confirm a clean chat / overlays disabled when possible.
  2. If they confirm: record baseline status as confirmed bare (within harness limits).
  3. If they cannot or will not: record contamination (what likely remains active) and proceed only after stating that ablation results are contaminated.
  4. Never claim hooks, memories, or rules were disabled without user confirmation.
  5. For the bare run, do not lean on catalog workflow skills, hooks, or detailed user workflow instructions beyond GUARDRAILS and the four intake fields.

Read the full file on GitHub · 138 lines

Files

What ships with it

3 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.

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. 5d ago First seen · 138 lines · 78 tokens per session scan A a01fde6ac942

Subscribe to this mod's changes

instruction-ablation is a skill published in the GitHub repository emaraschio/cursor-commands (9 stars, last pushed 1mo ago), licensed MIT. It adds 78 tokens to every session and 1,584 once invoked, about $0.0004 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 2026-09-03.